trl/examples/scripts/gkd.py at main · huggingface/trl · GitHub

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
# Copyright 2020-2026 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.

# /// script
# dependencies = [
#     "trl[peft]",
#     "trackio",
#     "kernels",
# ]
# ///

"""
# Full training:
python examples/scripts/gkd.py \
    --model_name_or_path Qwen/Qwen2-0.5B-Instruct \
    --teacher_model_name_or_path Qwen/Qwen2-1.5B-Instruct \
    --dataset_name trl-lib/chatbot_arena_completions \
    --learning_rate 2e-5 \
    --per_device_train_batch_size 4 \
    --gradient_accumulation_steps 8 \
    --output_dir gkd-model \
    --num_train_epochs 1 \
    --push_to_hub

# LoRA:
python examples/scripts/gkd.py \
    --model_name_or_path Qwen/Qwen2-0.5B-Instruct \
    --teacher_model_name_or_path Qwen/Qwen2-1.5B-Instruct \
    --dataset_name trl-lib/chatbot_arena_completions \
    --learning_rate 2e-4 \
    --per_device_train_batch_size 4 \
    --gradient_accumulation_steps 8 \
    --output_dir gkd-model \
    --num_train_epochs 1 \
    --push_to_hub \
    --use_peft \
    --lora_r 64 \
    --lora_alpha 16
"""

from datasets import load_dataset
from transformers import AutoTokenizer, GenerationConfig

from trl import (
    LogCompletionsCallback,
    ModelConfig,
    ScriptArguments,
    TrlParser,
    get_kbit_device_map,
    get_peft_config,
    get_quantization_config,
)
from trl.experimental.gkd import GKDConfig, GKDTrainer


if __name__ == "__main__":
    parser = TrlParser((ScriptArguments, GKDConfig, ModelConfig))
    script_args, training_args, model_args = parser.parse_args_and_config()

    ################
    # Model & Tokenizer
    ################
    model_kwargs = dict(
        revision=model_args.model_revision,
        attn_implementation=model_args.attn_implementation,
        dtype=model_args.dtype,
        use_cache=False if training_args.gradient_checkpointing else True,
    )
    quantization_config = get_quantization_config(model_args)
    if quantization_config is not None:
        # Passing None would not be treated the same as omitting the argument, so we include it only when valid.
        model_kwargs["device_map"] = get_kbit_device_map()
        model_kwargs["quantization_config"] = quantization_config

    training_args.model_init_kwargs = model_kwargs

    teacher_model_kwargs = dict(
        revision=model_args.model_revision,
        attn_implementation=model_args.attn_implementation,
        dtype=model_args.dtype,
        use_cache=True,
    )
    if quantization_config is not None:
        # Passing None would not be treated the same as omitting the argument, so we include it only when valid.
        model_kwargs["device_map"] = get_kbit_device_map()
        model_kwargs["quantization_config"] = quantization_config

    training_args.teacher_model_init_kwargs = teacher_model_kwargs

    tokenizer = AutoTokenizer.from_pretrained(
        model_args.model_name_or_path,
        revision=model_args.model_revision,
        padding_side="left",
    )
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token

    ################
    # Dataset
    ################
    dataset = load_dataset(script_args.dataset_name, name=script_args.dataset_config)

    ################
    # Training
    ################
    trainer = GKDTrainer(
        model=model_args.model_name_or_path,
        teacher_model=training_args.teacher_model_name_or_path,
        args=training_args,
        train_dataset=dataset[script_args.dataset_train_split],
        eval_dataset=dataset[script_args.dataset_test_split] if training_args.eval_strategy != "no" else None,
        processing_class=tokenizer,
        peft_config=get_peft_config(model_args),
    )

    if training_args.eval_strategy != "no":
        generation_config = GenerationConfig(
            max_new_tokens=training_args.max_new_tokens, do_sample=True, temperature=training_args.temperature
        )
        completions_callback = LogCompletionsCallback(trainer, generation_config, num_prompts=8)
        trainer.add_callback(completions_callback)

    trainer.train()

    # Save and push to hub
    trainer.save_model(training_args.output_dir)
    if training_args.push_to_hub:
        trainer.push_to_hub(dataset_name=script_args.dataset_name)