forked from NVIDIA/TensorRT-LLM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_precision_control.py
70 lines (59 loc) · 2.5 KB
/
test_precision_control.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
# 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 unittest
import numpy as np
import torch
from polygraphy.backend.trt import CreateConfig, EngineFromNetwork, TrtRunner
from transformers.models.llama.modeling_llama import LlamaRMSNorm
import tensorrt_llm
from tensorrt_llm import Tensor
from tensorrt_llm.functional import rms_norm
class TestPrecisionControl(unittest.TestCase):
def setUp(self):
tensorrt_llm.logger.set_level('error')
def test_precision_control(self):
# test data
test_shape = [2, 5, 10, 10]
dtype = 'float32'
x_data = torch.randn(*test_shape)
m = LlamaRMSNorm(test_shape[-1]) # LlamaRMSNorm only supports last dim
# construct trt network
builder = tensorrt_llm.Builder()
net = builder.create_network()
with tensorrt_llm.net_guard(net):
network = tensorrt_llm.default_trtnet()
x = Tensor(name='x',
shape=x_data.shape,
dtype=tensorrt_llm.str_dtype_to_trt(dtype))
output = rms_norm(x,
test_shape[-1],
weight=tensorrt_llm.constant(
m.weight.detach().cpu().numpy()))
output = output.trt_tensor
output.name = 'output'
network.mark_output(output)
# trt run
build_engine = EngineFromNetwork(
(builder.trt_builder, net.trt_network),
config=CreateConfig(precision_constraints='obey'))
with TrtRunner(build_engine) as runner:
outputs = runner.infer(feed_dict={'x': x_data.numpy()})
# pytorch run
with torch.no_grad():
ref = m(x_data)
# compare diff
np.testing.assert_allclose(ref.cpu().numpy(),
outputs['output'],
atol=1e-6)