mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-02-08 20:21:48 +08:00
119 lines
4.1 KiB
Python
119 lines
4.1 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.
|
|
import os
|
|
import sys
|
|
import unittest
|
|
|
|
import numpy as np
|
|
import pytest
|
|
import tensorrt as trt
|
|
import torch
|
|
from polygraphy.backend.trt import CreateConfig, EngineFromNetwork, TrtRunner
|
|
|
|
import tensorrt_llm
|
|
from tensorrt_llm import Tensor
|
|
from tensorrt_llm._utils import str_dtype_to_torch, str_dtype_to_trt
|
|
from tensorrt_llm.functional import low_latency_gemm
|
|
|
|
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
|
|
# Monkey Patching for torch.float8_e4m3fn support
|
|
from polygraphy.datatype import DataType
|
|
from utils.util import getSMVersion
|
|
|
|
original_to_dtype = DataType.to_dtype
|
|
|
|
|
|
def patched_to_dtype(dtype, target_module):
|
|
if dtype == DataType.FLOAT8E4M3FN and target_module == 'torch':
|
|
return torch.float8_e4m3fn
|
|
else:
|
|
return original_to_dtype(dtype, target_module)
|
|
|
|
|
|
DataType.to_dtype = patched_to_dtype
|
|
|
|
|
|
class TestLowLatencyGemm(unittest.TestCase):
|
|
|
|
def setUp(self) -> None:
|
|
tensorrt_llm.logger.set_level('error')
|
|
|
|
def reference_gemm_fp8(self, x, w, dtype):
|
|
w = w.transpose(0, 1).to(dtype=torch.float32)
|
|
y = torch.matmul(x.to(torch.float32), w)
|
|
return y.to(str_dtype_to_torch(dtype))
|
|
|
|
# float32
|
|
def run_low_latency_gemm_sm90(self, m, n, k, output_dtype):
|
|
torch.random.manual_seed(42)
|
|
shape_x = (m, k)
|
|
shape_w = (n, k)
|
|
x = torch.randint(-2, 2, shape_x).to(str_dtype_to_torch('fp8'))
|
|
w = torch.randint(-2, 2, shape_w).to(str_dtype_to_torch('fp8'))
|
|
# Create builder
|
|
builder = tensorrt_llm.Builder()
|
|
# Create empty network
|
|
net = builder.create_network()
|
|
net.plugin_config.low_latency_gemm_plugin = "fp8"
|
|
with tensorrt_llm.net_guard(net):
|
|
network = tensorrt_llm.default_trtnet()
|
|
# Init TensorRT-LLM tensor for x
|
|
x_tensor = Tensor(name='x',
|
|
shape=x.shape,
|
|
dtype=str_dtype_to_trt('fp8'))
|
|
# Init TensorRT-LLM tensor for w
|
|
w_tensor = Tensor(name='w',
|
|
shape=w.shape,
|
|
dtype=str_dtype_to_trt('fp8'))
|
|
# Get output tensor
|
|
output = low_latency_gemm(
|
|
x_tensor, w_tensor,
|
|
strict_dtype=str_dtype_to_trt(output_dtype)).trt_tensor
|
|
|
|
output.name = 'output'
|
|
network.mark_output(output)
|
|
output.dtype = str_dtype_to_trt(output_dtype)
|
|
|
|
engine = EngineFromNetwork(
|
|
(builder.trt_builder, net.trt_network),
|
|
config=CreateConfig(
|
|
fp8=True,
|
|
fp16=True,
|
|
memory_pool_limits={trt.MemoryPoolType.WORKSPACE: 33554432}))
|
|
|
|
feed_dict = {'x': x, "w": w}
|
|
with TrtRunner(engine) as runner:
|
|
outputs = runner.infer(feed_dict=feed_dict, check_inputs=False)
|
|
|
|
ref = self.reference_gemm_fp8(x, w, output_dtype)
|
|
np.testing.assert_allclose(ref.float().cpu().numpy(),
|
|
outputs['output'].float())
|
|
|
|
@pytest.mark.skipif(getSMVersion() != 90,
|
|
reason="LowLatencyGemm is only supported in SM90"
|
|
) # Skip tests that are not supported in SM90
|
|
def test_low_latency_gemm(self):
|
|
m = 64
|
|
n = 128
|
|
k = 128
|
|
output_dtype = "float32"
|
|
self.run_low_latency_gemm_sm90(m, n, k, output_dtype)
|
|
output_dtype = "float16"
|
|
self.run_low_latency_gemm_sm90(m, n, k, output_dtype)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|