Spaces:
Paused
Paused
File size: 9,175 Bytes
2f5127c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 |
# Copyright 2020-2025 The HuggingFace Team. 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 unittest
from datasets import load_dataset
from parameterized import parameterized
from transformers import AutoModelForCausalLM, AutoModelForSequenceClassification, AutoTokenizer
from transformers.testing_utils import require_peft
from transformers.utils import is_peft_available
from trl import NashMDConfig, NashMDTrainer
from .testing_utils import RandomPairwiseJudge, require_llm_blender
if is_peft_available():
from peft import LoraConfig, get_peft_model
class TestNashMDTrainer(unittest.TestCase):
def setUp(self):
self.model_id = "trl-internal-testing/tiny-Qwen2ForCausalLM-2.5"
self.model = AutoModelForCausalLM.from_pretrained(self.model_id)
self.ref_model = AutoModelForCausalLM.from_pretrained(self.model_id)
self.reward_model = AutoModelForSequenceClassification.from_pretrained(self.model_id, num_labels=1)
self.tokenizer = AutoTokenizer.from_pretrained(self.model_id)
self.tokenizer.pad_token = self.tokenizer.eos_token
@parameterized.expand([("standard_prompt_only",), ("conversational_prompt_only",)])
def test_nash_md_trainer_training(self, config_name):
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=2,
max_steps=3,
remove_unused_columns=False,
gradient_accumulation_steps=1,
learning_rate=9e-1,
eval_strategy="steps",
report_to="none",
)
dummy_dataset = load_dataset("trl-internal-testing/zen", config_name)
trainer = NashMDTrainer(
model=self.model,
ref_model=self.ref_model,
reward_model=self.reward_model,
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset["train"],
eval_dataset=dummy_dataset["test"],
)
trainer.train()
# Check if training loss is available
self.assertIn("train_loss", trainer.state.log_history[-1])
@require_peft
def test_training_with_peft(self):
lora_config = LoraConfig(r=16, lora_alpha=32, lora_dropout=0.05, bias="none", task_type="CAUSAL_LM")
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=2,
max_steps=3,
learning_rate=5.0e-7,
eval_strategy="steps",
report_to="none",
)
dummy_dataset = load_dataset("trl-internal-testing/zen", "standard_prompt_only")
trainer = NashMDTrainer(
model=self.model,
reward_model=self.reward_model,
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset["train"],
eval_dataset=dummy_dataset["test"],
peft_config=lora_config,
)
trainer.train()
# Check if training loss is available
self.assertIn("train_loss", trainer.state.log_history[-1])
@require_peft
def test_training_with_peft_and_ref_model(self):
lora_config = LoraConfig(r=16, lora_alpha=32, lora_dropout=0.05, bias="none", task_type="CAUSAL_LM")
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=2,
max_steps=3,
learning_rate=5.0e-7,
eval_strategy="steps",
report_to="none",
)
dummy_dataset = load_dataset("trl-internal-testing/zen", "standard_prompt_only")
trainer = NashMDTrainer(
model=self.model,
ref_model=self.ref_model,
reward_model=self.reward_model,
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset["train"],
eval_dataset=dummy_dataset["test"],
peft_config=lora_config,
)
trainer.train()
# Check if training loss is available
self.assertIn("train_loss", trainer.state.log_history[-1])
@require_peft
def test_training_with_peft_model_and_peft_config(self):
model_lora_config = LoraConfig(r=8, lora_alpha=16, lora_dropout=0.1, bias="none", task_type="CAUSAL_LM")
model = get_peft_model(self.model, model_lora_config)
# we want only the "train adapter" to be trained
lora_train_config = LoraConfig(r=16, lora_alpha=32, lora_dropout=0.05, bias="none", task_type="CAUSAL_LM")
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=2,
max_steps=3,
learning_rate=5.0e-7,
eval_strategy="steps",
report_to="none",
)
dummy_dataset = load_dataset("trl-internal-testing/zen", "standard_prompt_only")
trainer = NashMDTrainer(
model=model,
reward_model=self.reward_model,
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset["train"],
eval_dataset=dummy_dataset["test"],
peft_config=lora_train_config,
)
trainer.train()
# Check if training loss is available
self.assertIn("train_loss", trainer.state.log_history[-1])
@require_peft
def test_training_pre_pefted_model_implicit_ref_with_reward_model(self):
lora_config = LoraConfig(r=8, lora_alpha=16, lora_dropout=0.1, bias="none", task_type="CAUSAL_LM")
# self.model from setUp is a base AutoModelForCausalLM
peft_model_instance = get_peft_model(self.model, lora_config)
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=1, # Keep small for quick test
max_steps=2, # Few steps
learning_rate=5.0e-7,
eval_strategy="no",
report_to="none",
remove_unused_columns=False, # Important for the dummy dataset
)
dummy_dataset = load_dataset("trl-internal-testing/zen", "standard_prompt_only")["train"]
trainer = NashMDTrainer(
model=peft_model_instance, # Pass the already PEFT model
ref_model=None, # Implicit reference from peft_model_instance's base
reward_model=self.reward_model, # To trigger GeometricMixtureWrapper path
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset,
# peft_config is not passed, as model is already PEFT
)
trainer.train()
self.assertIn("train_loss", trainer.state.log_history[-1])
@parameterized.expand([("standard_prompt_only",), ("conversational_prompt_only",)])
@require_llm_blender
def test_nash_md_trainer_judge_training(self, config_name):
with tempfile.TemporaryDirectory() as tmp_dir:
training_args = NashMDConfig(
output_dir=tmp_dir,
per_device_train_batch_size=2,
max_steps=3,
remove_unused_columns=False,
gradient_accumulation_steps=1,
learning_rate=9e-1,
eval_strategy="steps",
report_to="none",
)
dummy_dataset = load_dataset("trl-internal-testing/zen", config_name)
judge = RandomPairwiseJudge()
trainer = NashMDTrainer(
model=self.model,
ref_model=self.ref_model,
judge=judge,
args=training_args,
processing_class=self.tokenizer,
train_dataset=dummy_dataset["train"],
eval_dataset=dummy_dataset["test"],
)
trainer.train()
# Check if training loss is available
self.assertIn("train_loss", trainer.state.log_history[-1])
|