File size: 8,051 Bytes
b386992 | 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 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 | # Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
#
# 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 tempfile
import nemo_run as run
import pytest
import torch
from nemo import lightning as nl
from nemo.collections import llm
from nemo.collections.llm.api import _validate_config
from nemo.collections.llm.gpt.model.llama import Llama3Config8B, LlamaModel
class TestValidateConfig:
def reset_configs(self):
model = LlamaModel(config=run.Config(Llama3Config8B))
data = llm.MockDataModule(seq_length=4096, global_batch_size=16, micro_batch_size=2)
trainer = nl.Trainer(strategy=nl.MegatronStrategy())
return model, data, trainer
def test_model_validation(self):
model, data, trainer = self.reset_configs()
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.seq_length = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.num_layers = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.hidden_size = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.num_attention_heads = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.ffn_hidden_size = 0
_validate_config(model, data, trainer)
def test_data_validation(self):
model, data, trainer = self.reset_configs()
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
data.micro_batch_size = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
data.global_batch_size = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
data.seq_length = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
data.micro_batch_size = 3
data.global_batch_size = 128
_validate_config(model, data, trainer)
def test_trainer_validatiopn(self):
model, data, trainer = self.reset_configs()
_validate_config(model, data, trainer)
# Basic validation
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.tensor_model_parallel_size = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.pipeline_model_parallel_size = 0
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.context_parallel_size = 0
_validate_config(model, data, trainer)
# DP validation
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.tensor_model_parallel_size = 8
trainer.strategy.pipeline_model_parallel_size = 2
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.tensor_model_parallel_size = 3
trainer.strategy.pipeline_model_parallel_size = 2
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
data.global_batch_size = 3
data.micro_batch_size = 1
trainer.strategy.tensor_model_parallel_size = 2
trainer.strategy.pipeline_model_parallel_size = 2
_validate_config(model, data, trainer)
# TP/SP validation
model, data, trainer = self.reset_configs()
trainer.strategy.tensor_model_parallel_size = 1
trainer.strategy.sequence_parallel = True
_validate_config(model, data, trainer)
assert trainer.strategy.sequence_parallel == False
# PP/VP validation
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
trainer.strategy.pipeline_model_parallel_size = 2
trainer.strategy.pipeline_dtype = None
_validate_config(model, data, trainer)
model, data, trainer = self.reset_configs()
trainer.strategy.pipeline_model_parallel_size = 1
trainer.strategy.virtual_pipeline_model_parallel_size = 2
trainer.strategy.pipeline_dtype = torch.bfloat16
_validate_config(model, data, trainer)
assert trainer.strategy.virtual_pipeline_model_parallel_size is None
assert trainer.strategy.pipeline_dtype is None
# CP validation
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.seq_length = 5
trainer.strategy.context_parallel_size = 2
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.seq_length = 2
trainer.strategy.context_parallel_size = 2
_validate_config(model, data, trainer)
# EP validation
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.num_moe_experts = None
trainer.strategy.expert_model_parallel_size = 2
_validate_config(model, data, trainer)
with pytest.raises(AssertionError):
model, data, trainer = self.reset_configs()
model.config.num_moe_experts = 3
trainer.strategy.expert_model_parallel_size = 2
_validate_config(model, data, trainer)
class TestImportCkpt:
def test_output_path_exists_no_overwrite(self):
"""Test that an error is raised when the output path exists and overwrite is set to False."""
with pytest.raises(FileExistsError), tempfile.TemporaryDirectory() as output_path:
llm.import_ckpt(
model=llm.LlamaModel(config=llm.Llama32Config1B()),
source="hf://meta-llama/Llama-3.2-1B",
output_path=output_path,
overwrite=False,
)
class TestExportCkpt:
def test_output_path_exists_no_overwrite(self):
"""Test that an error is raised when the output path exists and overwrite is set to False."""
with (
pytest.raises(FileExistsError),
tempfile.TemporaryDirectory() as output_path,
tempfile.TemporaryDirectory() as path,
):
llm.export_ckpt(
path=path,
target="hf",
output_path=output_path,
overwrite=False,
)
|