Changes to TRT-LLM download tool for multigpu distributed case

.github/scripts/filter-matrix.py

@@ -60,6 +60,41 @@ def filter_matrix_item(
         return True
 
 
+def create_distributed_config(item: Dict[str, Any]) -> Dict[str, Any]:
+    """Create distributed test configuration from a regular config.
+
+    Takes a standard test config and modifies it for distributed testing:
+    - Changes runner to multi-GPU instance
+    - Adds num_gpus field
+    - Adds config marker
+    """
+    import sys
+
+    # Create a copy to avoid modifying the original
+    dist_item = item.copy()
+
+    # Debug: Show original config
+    print(f"[DEBUG] Creating distributed config from:", file=sys.stderr)
+    print(f"[DEBUG]   Python: {item.get('python_version')}", file=sys.stderr)
+    print(f"[DEBUG]   CUDA: {item.get('desired_cuda')}", file=sys.stderr)
+    print(
+        f"[DEBUG]   Original runner: {item.get('validation_runner')}", file=sys.stderr
+    )
+
+    # Override runner to use multi-GPU instance
+    dist_item["validation_runner"] = "linux.g4dn.12xlarge.nvidia.gpu"
+
+    # Add distributed-specific fields
+    dist_item["num_gpus"] = 2
+    dist_item["config"] = "distributed"
+
+    # Debug: Show modified config
+    print(f"[DEBUG]   New runner: {dist_item['validation_runner']}", file=sys.stderr)
+    print(f"[DEBUG]   GPUs: {dist_item['num_gpus']}", file=sys.stderr)
+
+    return dist_item
+
+
 def main(args: list[str]) -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument(
@@ -99,16 +134,69 @@ def main(args: list[str]) -> None:
 
     includes = matrix_dict["include"]
     filtered_includes = []
+    distributed_includes = []  # NEW: separate list for distributed configs
+
+    print(f"[DEBUG] Processing {len(includes)} input configs", file=sys.stderr)
 
     for item in includes:
+        py_ver = item.get("python_version", "unknown")
+        cuda_ver = item.get("desired_cuda", "unknown")
+
+        print(f"[DEBUG] Checking config: py={py_ver}, cuda={cuda_ver}", file=sys.stderr)
+
         if filter_matrix_item(
             item,
             options.jetpack == "true",
             options.limit_pr_builds == "true",
         ):
+            print(f"[DEBUG] passed filter - adding to build matrix", file=sys.stderr)
             filtered_includes.append(item)
 
-    filtered_matrix_dict = {"include": filtered_includes}
+            # NEW: Create distributed variant for specific configs
+            # Only Python 3.10 + CUDA 13.0 for now
+            if item["python_version"] == "3.10" and item["desired_cuda"] == "cu130":
+                print(
+                    f"[DEBUG]  Creating distributed config for py3.10+cu130",
+                    file=sys.stderr,
+                )
+                distributed_includes.append(create_distributed_config(item))
+        else:
+            print(f"[DEBUG] FILTERED OUT", file=sys.stderr)
+
+    # Debug: Show summary
+    print(f"[DEBUG] Final counts:", file=sys.stderr)
+    print(f"[DEBUG]   Regular configs: {len(filtered_includes)}", file=sys.stderr)
+    print(
+        f"[DEBUG]   Distributed configs: {len(distributed_includes)}", file=sys.stderr
+    )
+
+    # Debug: Show which configs will be built
+    print(
+        f"[DEBUG] Configs that will be BUILT (in filtered_includes):", file=sys.stderr
+    )
+    for item in filtered_includes:
+        print(
+            f"[DEBUG]   - py={item.get('python_version')}, cuda={item.get('desired_cuda')}",
+            file=sys.stderr,
+        )
+
+    print(
+        f"[DEBUG] Configs for DISTRIBUTED TESTS (in distributed_includes):",
+        file=sys.stderr,
+    )
+    for item in distributed_includes:
+        print(
+            f"[DEBUG]   - py={item.get('python_version')}, cuda={item.get('desired_cuda')}, gpus={item.get('num_gpus')}",
+            file=sys.stderr,
+        )
+
+    # NEW: Output both regular and distributed configs
+    filtered_matrix_dict = {
+        "include": filtered_includes,
+        "distributed_include": distributed_includes,  # NEW field
+    }
+
+    # Output to stdout (consumed by GitHub Actions)
     print(json.dumps(filtered_matrix_dict))
 
 

.github/workflows/build-test-linux-x86_64.yml

@@ -68,7 +68,11 @@ jobs:
       ref: ""
       test-infra-repository: pytorch/test-infra
       test-infra-ref: main
-      build-matrix: ${{ needs.filter-matrix.outputs.matrix }}
+      # Extract the include array from filter-matrix output
+      build-matrix: |
+        {
+          "include": ${{ toJSON(fromJSON(needs.filter-matrix.outputs.matrix).include) }}
+        }
       pre-script: ${{ matrix.pre-script }}
       env-var-script: ${{ matrix.env-var-script }}
       post-script: ${{ matrix.post-script }}
@@ -480,18 +484,50 @@ jobs:
       ref: ""
       test-infra-repository: pytorch/test-infra
       test-infra-ref: main
-      build-matrix: ${{ needs.filter-matrix.outputs.matrix }}
+      # Extract the distributed_include array from filter-matrix output
+      build-matrix: |
+        {
+          "include": ${{ toJSON(fromJSON(needs.filter-matrix.outputs.matrix).distributed_include) }}
+        }
       pre-script: ${{ matrix.pre-script }}
       script: |
         set -euo pipefail
+
+        # Debug: Show what config we're using
+        echo "=========================================="
+        echo "DISTRIBUTED TEST CONFIGURATION"
+        echo "=========================================="
+        echo "Python version: ${PYTHON_VERSION}"
+        echo "CUDA version: ${CU_VERSION}"
+        echo "Runner: ${{ matrix.validation_runner }}"
+        echo "Num GPUs: ${{ matrix.num_gpus }}"
+        echo "Config: ${{ matrix.config }}"
+        echo "=========================================="
+
+        # Verify GPUs are available
+        echo "Checking GPU availability:"
+        nvidia-smi
+        echo "GPU count: $(nvidia-smi --query-gpu=name --format=csv,noheader | wc -l)"
+        echo "=========================================="
+
         export USE_HOST_DEPS=1
         export CI_BUILD=1
         export USE_TRTLLM_PLUGINS=1
+
+        # Install MPI (required for TensorRT-LLM plugins)
+        echo "Installing MPI..."
         dnf install -y mpich mpich-devel openmpi openmpi-devel
+
+        # Run distributed tests
         pushd .
-        cd tests/py
-        cd dynamo
-        python -m pytest -ra --junitxml=${RUNNER_TEST_RESULTS_DIR}/l2_dynamo_distributed_test_results.xml distributed/test_nccl_ops.py
+        cd tests/py/dynamo
+
+        echo "Running distributed tests with mpirun..."
+        mpirun --allow-run-as-root -n ${{ matrix.num_gpus }} \
+          python -m pytest -ra \
+          --junitxml=${RUNNER_TEST_RESULTS_DIR}/l2_dynamo_distributed_test_results.xml \
+          distributed/test_nccl_ops.py
+
         popd
 
 concurrency:

.github/workflows/build_linux.yml

@@ -160,6 +160,17 @@ jobs:
       options: ${{ matrix.gpu_arch_type == 'cuda' && '--gpus all' || ' ' }}
     timeout-minutes: ${{ inputs.timeout }}
     steps:
+      - name: Debug matrix configuration
+        shell: bash
+        run: |
+          echo "=========================================="
+          echo "BUILD MATRIX DEBUG"
+          echo "=========================================="
+          echo "Python version: ${{ matrix.python_version }}"
+          echo "CUDA version: ${{ matrix.desired_cuda }}"
+          echo "GPU arch type: ${{ matrix.gpu_arch_type }}"
+          echo "Runner: ${{ matrix.validation_runner }}"
+          echo "=========================================="
       - name: Clean workspace
         shell: bash -l {0}
         run: |

examples/distributed_inference/tensor_parallel_initialize_dist.py

@@ -14,32 +14,13 @@
 import tensorrt as trt
 import torch
 import torch.distributed as dist
-from torch.distributed._tensor.device_mesh import init_device_mesh
+from torch.distributed._tensor.device_mesh import DeviceMesh, init_device_mesh
 
+logger = logging.getLogger(__name__)
 
-def find_repo_root(max_depth=10):
-    dir_path = os.path.dirname(os.path.realpath(__file__))
-    for i in range(max_depth):
-        files = os.listdir(dir_path)
-        if "MODULE.bazel" in files:
-            return dir_path
-        else:
-            dir_path = os.path.dirname(dir_path)
 
-    raise RuntimeError("Could not find repo root")
-
-
-def initialize_logger(rank, logger_file_name):
-    logger = logging.getLogger()
-    logger.setLevel(logging.INFO)
-    fh = logging.FileHandler(logger_file_name + f"_{rank}.log", mode="w")
-    fh.setLevel(logging.INFO)
-    logger.addHandler(fh)
-    return logger
-
-
-# This is required for env initialization since we use mpirun
-def initialize_distributed_env(logger_file_name, rank=0, world_size=1, port=29500):
+# this is kept at the application level, when mpirun is used to run the application
+def initialize_distributed_env(rank=0, world_size=1, port=29500):
     local_rank = int(
         os.environ.get("OMPI_COMM_WORLD_LOCAL_RANK", rank % torch.cuda.device_count())
     )
@@ -50,9 +31,6 @@ def initialize_distributed_env(logger_file_name, rank=0, world_size=1, port=2950
     os.environ["WORLD_SIZE"] = str(world_size)
     os.environ["MASTER_ADDR"] = "127.0.0.1"
     os.environ["MASTER_PORT"] = str(port)
-    os.environ["TRTLLM_PLUGINS_PATH"] = (
-        find_repo_root() + "/lib/libnvinfer_plugin_tensorrt_llm.so"
-    )
 
     # Necessary to assign a device to each rank.
     torch.cuda.set_device(local_rank)
@@ -66,16 +44,49 @@ def initialize_distributed_env(logger_file_name, rank=0, world_size=1, port=2950
     device_mesh = init_device_mesh(device_type="cuda", mesh_shape=(world_size,))
     rank = device_mesh.get_rank()
     assert rank == local_rank
-    logger = initialize_logger(rank, logger_file_name)
     device_id = (
         rank % torch.cuda.device_count()
     )  # Ensure each rank gets a unique device
     torch.cuda.set_device(device_id)
 
-    return device_mesh, world_size, rank, logger
+    return device_mesh, world_size, rank
 
 
 def cleanup_distributed_env():
     """Clean up distributed process group to prevent resource leaks."""
     if dist.is_initialized():
         dist.destroy_process_group()
+
+
+def check_tensor_parallel_device_number(world_size: int) -> None:
+    if world_size % 2 != 0:
+        raise ValueError(
+            f"TP examples require even number of GPUs, but got {world_size} gpus"
+        )
+
+
+def get_tensor_parallel_device_mesh(
+    rank: int = 0, world_size: int = 1
+) -> tuple[DeviceMesh, int, int]:
+    local_rank = int(
+        os.environ.get("OMPI_COMM_WORLD_LOCAL_RANK", rank % torch.cuda.device_count())
+    )
+    world_size = int(os.environ.get("OMPI_COMM_WORLD_SIZE", world_size))
+    device_mesh = init_device_mesh(device_type="cuda", mesh_shape=(world_size,))
+    rank = device_mesh.get_rank()
+    assert rank == local_rank
+    device_id = (
+        rank % torch.cuda.device_count()
+    )  # Ensure each rank gets a unique device
+    torch.cuda.set_device(device_id)
+
+    return device_mesh, world_size, rank
+
+
+def initialize_distributed_logger(rank: int, logger_file_name: str) -> logging.Logger:
+    logger = logging.getLogger()
+    logger.setLevel(logging.INFO)
+    fh = logging.FileHandler(logger_file_name + f"_{rank}.log", mode="w")
+    fh.setLevel(logging.INFO)
+    logger.addHandler(fh)
+    return logger

examples/distributed_inference/tensor_parallel_rotary_embedding.py

@@ -9,26 +9,31 @@
 
 """
 
-import logging
-import os
 import time
 
 import torch
-import torch_tensorrt
-from rotary_embedding import RotaryAttention, parallel_rotary_block
+import torch.distributed as dist
 from tensor_parallel_initialize_dist import (
     cleanup_distributed_env,
+    get_tensor_parallel_device_mesh,
     initialize_distributed_env,
+    initialize_distributed_logger,
 )
 
-device_mesh, _world_size, _rank, logger = initialize_distributed_env(
-    "./tensor_parallel_rotary_embedding"
-)
+if not dist.is_initialized():
+    initialize_distributed_env()
 
+import torch_tensorrt
+
+device_mesh, _world_size, _rank = get_tensor_parallel_device_mesh()
+logger = initialize_distributed_logger(_rank, "tensor_parallel_rotary_embedding")
+
+from rotary_embedding import RotaryAttention, parallel_rotary_block
 
 """
 This example covers the rotary embedding in Llama3 model and is derived from https://lightning.ai/lightning-ai/studios/tensor-parallelism-supercharging-large-model-training-with-pytorch-lightning
-Command to run with single GPU: mpirun -n 1 --allow-run-as-root python tensor_parallel_rotary_embedding.py
+Command to run with single GPU: USE_TRTLLM_PLUGINS=1 mpirun -n 1 --allow-run-as-root python tensor_parallel_rotary_embedding.py
+Command to run with 2 GPUs: USE_TRTLLM_PLUGINS=1 mpirun -n 2 --allow-run-as-root python tensor_parallel_rotary_embedding.py
 """
 
 BATCH = 2

examples/distributed_inference/tensor_parallel_simple_example.py

@@ -16,7 +16,7 @@
 -----
 .. code-block:: bash
 
-    mpirun -n 2 --allow-run-as-root python tensor_parallel_simple_example.py
+    USE_TRTLLM_PLUGINS=1 mpirun -n 2 --allow-run-as-root python tensor_parallel_simple_example.py
 """
 
 import time
@@ -25,22 +25,31 @@
 import torch
 import torch.distributed as dist
 import torch.nn as nn
-import torch_tensorrt
 from tensor_parallel_initialize_dist import (
     cleanup_distributed_env,
+    get_tensor_parallel_device_mesh,
     initialize_distributed_env,
+    initialize_distributed_logger,
 )
+
+if not dist.is_initialized():
+    initialize_distributed_env()
+import torch_tensorrt
 from torch.distributed._tensor import Shard
 from torch.distributed.tensor.parallel import (
     ColwiseParallel,
     RowwiseParallel,
     parallelize_module,
 )
-
-device_mesh, _world_size, _rank, logger = initialize_distributed_env(
-    "./tensor_parallel_simple_example"
+from torch_tensorrt.dynamo.distributed.utils import (
+    get_tensor_parallel_device_mesh,
+    initialize_distributed_logger,
 )
 
+device_mesh, _world_size, _rank = get_tensor_parallel_device_mesh()
+logger = initialize_distributed_logger(_rank, "tensor_parallel_simple_example")
+
+
 """
 This example takes some code from https://github.com/pytorch/examples/blob/main/distributed/tensor_parallelism/tensor_parallel_example.py
 """