TensorRT-LLMs/tests/model/redrafter/test_top1.py
2024-08-29 17:25:07 +08:00

83 lines
2.8 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 torch
import tensorrt_llm
import tensorrt_llm.models.redrafter
import tensorrt_llm.models.redrafter.redrafter_helper
from tensorrt_llm import Tensor
sys.path.append(os.path.join(os.path.dirname(__file__), os.pardir, os.pardir))
from utils.util import create_session, run_session
NINF = -50000.0
class TestReDrafter(unittest.TestCase):
def setUp(self):
tensorrt_llm.logger.set_level('warning')
########################################################################################################################
def test_top_1_logits(self):
# test data
bs = 2
S = 5
V = 4
old_device = torch.get_default_device()
torch.set_default_device("cuda")
torch.manual_seed(0)
logits = torch.rand((bs, S, V), dtype=torch.float32)
ref_res = torch.tensor(
[[[NINF, NINF, NINF, -0.], [-0., NINF, NINF, NINF],
[NINF, -0., NINF, NINF], [NINF, -0., NINF, NINF],
[-0., NINF, NINF, NINF]],
[[-0., NINF, NINF, NINF], [NINF, -0., NINF, NINF],
[-0., NINF, NINF, NINF], [NINF, -0., NINF, NINF],
[NINF, -0., NINF, NINF]]],
dtype=torch.float32)
# construct trt network
builder = tensorrt_llm.Builder()
network = builder.create_network()
with tensorrt_llm.net_guard(network):
logits_t = Tensor(name='logits',
shape=logits.shape,
dtype=tensorrt_llm.torch_dtype_to_trt(
logits.dtype))
outputs = tensorrt_llm.models.redrafter.redrafter_helper._top_1_logits(
logits_t, NINF)
outputs.mark_output("outputs")
# trt run
session = create_session(
builder,
network,
precision='float32',
)
inputs = {
'logits': logits,
}
outputs = run_session(session, inputs)
torch.testing.assert_close(outputs['outputs'], ref_res, rtol=0, atol=0)
torch.set_default_device(old_device)
return