mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-22 19:52:38 +08:00
111 lines
3.9 KiB
Python
111 lines
3.9 KiB
Python
# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
# SPDX-License-Identifier: Apache-2.0
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
"""Module test_dit test dit examples."""
|
|
|
|
import os
|
|
|
|
import pytest
|
|
from defs.common import convert_weights, venv_check_call, venv_mpi_check_call
|
|
from defs.conftest import get_device_count, skip_fp8_pre_ada
|
|
from defs.trt_test_alternative import check_call, exists
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def dit_example_root(llm_root):
|
|
"Get DiT example root"
|
|
example_root = os.path.join(llm_root, "examples", "dit")
|
|
|
|
return example_root
|
|
|
|
|
|
@pytest.mark.skip_less_device_memory(50000)
|
|
@pytest.mark.parametrize("tp_size", [1, 4], ids=lambda tp_size: f'tp{tp_size}')
|
|
@pytest.mark.parametrize(
|
|
"llm_dit_model_root",
|
|
["dit-xl-2-256x256", "dit-xl-2-512x512", "dit-xl-2-512x512-fp8-linear"],
|
|
indirect=True)
|
|
def test_llm_dit_multiple_gpus(dit_example_root, llm_dit_model_root, llm_venv,
|
|
engine_dir, cmodel_dir, tp_size):
|
|
"Build & run dit."
|
|
if get_device_count() < tp_size:
|
|
pytest.skip(f"Device number is less than {tp_size}")
|
|
|
|
skip_fp8_pre_ada("fp8" in llm_dit_model_root.lower())
|
|
|
|
workspace = llm_venv.get_working_directory()
|
|
dtype = "float16"
|
|
tp_size, pp_size = tp_size, 1
|
|
world_size = tp_size * pp_size
|
|
model_name = os.path.basename(llm_dit_model_root)
|
|
image_size = 512 if "512" in llm_dit_model_root else 256
|
|
input_size = 64 if image_size == 512 else 32
|
|
onnx_file = os.path.join(workspace, "vae_decoder/onnx/visual_encoder.onnx")
|
|
plan_file = os.path.join(workspace,
|
|
"vae_decoder/plan/visual_encoder_fp16.plan")
|
|
|
|
enable_fp8_linear = True if "FP8" in llm_dit_model_root else False
|
|
print("Convert weight...")
|
|
model_dir = convert_weights(llm_venv=llm_venv,
|
|
example_root=dit_example_root,
|
|
cmodel_dir=cmodel_dir,
|
|
model=model_name,
|
|
model_path=llm_dit_model_root,
|
|
data_type=dtype,
|
|
tp_size=tp_size,
|
|
pp_size=pp_size,
|
|
input_size=input_size,
|
|
fp8_linear=enable_fp8_linear)
|
|
|
|
print("Build TRT-LLM engines...")
|
|
build_cmd = [
|
|
"trtllm-build",
|
|
f"--checkpoint_dir={model_dir}",
|
|
f"--output_dir={engine_dir}",
|
|
f"--workers={world_size}",
|
|
"--max_batch_size=8",
|
|
"--remove_input_padding=disable",
|
|
"--bert_attention_plugin=disable",
|
|
]
|
|
|
|
check_call(" ".join(build_cmd), shell=True, env=llm_venv._new_env)
|
|
|
|
print("Build VAE engines...")
|
|
build_vae_cmd = [
|
|
f"{dit_example_root}/vae_decoder_trt.py",
|
|
f"--image-size={image_size}",
|
|
f"--onnxFile={onnx_file}",
|
|
f"--planFile={plan_file}",
|
|
"--max_batch_size=8",
|
|
]
|
|
|
|
venv_check_call(llm_venv, build_vae_cmd)
|
|
|
|
print("Run summary...")
|
|
run_cmd = [
|
|
f"{dit_example_root}/sample.py",
|
|
f"--vae_decoder_engine={plan_file}",
|
|
f"--tllm_model_dir={engine_dir}",
|
|
f"--image-size={image_size}",
|
|
]
|
|
|
|
if world_size > 1:
|
|
venv_mpi_check_call(
|
|
llm_venv, ["mpirun", "-n", f"{world_size}", "--allow-run-as-root"],
|
|
run_cmd)
|
|
else:
|
|
venv_check_call(llm_venv, run_cmd)
|
|
|
|
assert exists(f"{workspace}/sample.png")
|