File size: 5,594 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
# 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 shutil

from accelerate import PartialState
from datasets import load_dataset
from transformers import (
    AutoModelForCausalLM,
    AutoModelForSequenceClassification,
    AutoTokenizer,
    HfArgumentParser,
)

from trl import ModelConfig, RLOOConfig, RLOOTrainer, ScriptArguments
from trl.trainer.utils import SIMPLE_CHAT_TEMPLATE


"""
python examples/scripts/rloo/rloo_tldr.py \
    --dataset_name trl-internal-testing/tldr-preference-sft-trl-style \
    --dataset_test_split validation \
    --learning_rate 3e-6 \
    --output_dir models/minimal/ppo \
    --per_device_train_batch_size 1 \
    --gradient_accumulation_steps 64 \
    --total_episodes 30000 \
    --model_name_or_path EleutherAI/pythia-1b-deduped \
    --sft_model_path cleanrl/EleutherAI_pythia-1b-deduped__sft__tldr \
    --reward_model_path cleanrl/EleutherAI_pythia-1b-deduped__reward__tldr \
    --missing_eos_penalty 1.0 \
    --stop_token eos \
    --response_length 53

accelerate launch --config_file examples/accelerate_configs/deepspeed_zero2.yaml \
    examples/scripts/rloo/rloo_tldr.py \
    --dataset_name trl-internal-testing/tldr-preference-sft-trl-style \
    --dataset_test_split validation \
    --output_dir models/minimal/rloo_tldr \
    --num_ppo_epochs 1 \
    --num_mini_batches 1 \
    --learning_rate 3e-6 \
    --per_device_train_batch_size 16 \
    --gradient_accumulation_steps 4 \
    --total_episodes 1000000 \
    --model_name_or_path EleutherAI/pythia-1b-deduped \
    --sft_model_path cleanrl/EleutherAI_pythia-1b-deduped__sft__tldr \
    --reward_model_path cleanrl/EleutherAI_pythia-1b-deduped__reward__tldr \
    --local_rollout_forward_batch_size 16 \
    --missing_eos_penalty 1.0 \
    --stop_token eos
"""


if __name__ == "__main__":
    parser = HfArgumentParser((ScriptArguments, RLOOConfig, ModelConfig))
    script_args, training_args, model_args = parser.parse_args_into_dataclasses()
    # remove output_dir if exists
    shutil.rmtree(training_args.output_dir, ignore_errors=True)

    ################
    # Model & Tokenizer
    ################
    tokenizer = AutoTokenizer.from_pretrained(
        model_args.model_name_or_path, padding_side="left", trust_remote_code=model_args.trust_remote_code
    )
    tokenizer.add_special_tokens({"pad_token": "[PAD]"})
    if tokenizer.chat_template is None:
        tokenizer.chat_template = SIMPLE_CHAT_TEMPLATE
    reward_model = AutoModelForSequenceClassification.from_pretrained(
        training_args.reward_model_path, trust_remote_code=model_args.trust_remote_code, num_labels=1
    )
    ref_policy = AutoModelForCausalLM.from_pretrained(
        training_args.sft_model_path, trust_remote_code=model_args.trust_remote_code
    )
    policy = AutoModelForCausalLM.from_pretrained(
        training_args.sft_model_path, trust_remote_code=model_args.trust_remote_code
    )
    ################
    # Dataset
    ################
    dataset = load_dataset(script_args.dataset_name, name=script_args.dataset_config)
    train_dataset = dataset[script_args.dataset_train_split]
    eval_dataset = dataset[script_args.dataset_test_split] if training_args.eval_strategy != "no" else None

    def prepare_dataset(dataset, tokenizer):
        """pre-tokenize the dataset before training; only collate during training"""

        def tokenize(element):
            input_ids = tokenizer.apply_chat_template(
                element["messages"][:1],
                padding=False,
                add_generation_prompt=True,
            )
            return {"input_ids": input_ids, "lengths": len(input_ids)}

        return dataset.map(
            tokenize,
            remove_columns=dataset.column_names,
            num_proc=training_args.dataset_num_proc,
        )

    # Compute that only on the main process for faster data processing.
    # see: https://github.com/huggingface/trl/pull/1255
    with PartialState().local_main_process_first():
        train_dataset = prepare_dataset(train_dataset, tokenizer)
        eval_dataset = prepare_dataset(eval_dataset, tokenizer)
        # filtering
        train_dataset = train_dataset.filter(lambda x: x["lengths"] <= 512, num_proc=training_args.dataset_num_proc)
        eval_dataset = eval_dataset.filter(lambda x: x["lengths"] <= 512, num_proc=training_args.dataset_num_proc)

    assert train_dataset[0]["input_ids"][-1] != tokenizer.eos_token_id, "The last token should not be an EOS token"
    ################
    # Training
    ################
    trainer = RLOOTrainer(
        config=training_args,
        processing_class=tokenizer,
        policy=policy,
        ref_policy=ref_policy,
        reward_model=reward_model,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
    )
    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)

    trainer.generate_completions()