Spaces:
Paused
Paused
File size: 5,993 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 |
# 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
import torch
from accelerate import PartialState
from datasets import load_dataset
from transformers import (
AutoModelForCausalLM,
AutoModelForSequenceClassification,
AutoTokenizer,
HfArgumentParser,
)
from trl import (
ModelConfig,
PPOConfig,
PPOTrainer,
ScriptArguments,
get_kbit_device_map,
get_peft_config,
get_quantization_config,
)
from trl.trainer.utils import SIMPLE_CHAT_TEMPLATE
"""
python -i examples/scripts/ppo/ppo.py \
--dataset_name trl-internal-testing/descriptiveness-sentiment-trl-style \
--dataset_train_split descriptiveness \
--learning_rate 3e-6 \
--output_dir models/minimal/ppo \
--per_device_train_batch_size 64 \
--gradient_accumulation_steps 1 \
--total_episodes 10000 \
--model_name_or_path EleutherAI/pythia-1b-deduped \
--missing_eos_penalty 1.0
accelerate launch --config_file examples/accelerate_configs/deepspeed_zero3.yaml \
examples/scripts/ppo/ppo.py \
--dataset_name trl-internal-testing/descriptiveness-sentiment-trl-style \
--dataset_train_split descriptiveness \
--output_dir models/minimal/ppo \
--num_ppo_epochs 1 \
--num_mini_batches 1 \
--learning_rate 3e-6 \
--per_device_train_batch_size 1 \
--gradient_accumulation_steps 16 \
--total_episodes 10000 \
--model_name_or_path EleutherAI/pythia-1b-deduped \
--sft_model_path EleutherAI/pythia-1b-deduped \
--reward_model_path EleutherAI/pythia-1b-deduped \
--local_rollout_forward_batch_size 1 \
--missing_eos_penalty 1.0
"""
if __name__ == "__main__":
parser = HfArgumentParser((ScriptArguments, PPOConfig, 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
################
torch_dtype = (
model_args.torch_dtype if model_args.torch_dtype in ["auto", None] else getattr(torch, model_args.torch_dtype)
)
quantization_config = get_quantization_config(model_args)
model_kwargs = dict(
revision=model_args.model_revision,
attn_implementation=model_args.attn_implementation,
torch_dtype=torch_dtype,
device_map=get_kbit_device_map() if quantization_config is not None else None,
quantization_config=quantization_config,
)
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
value_model = AutoModelForSequenceClassification.from_pretrained(
training_args.reward_model_path, trust_remote_code=model_args.trust_remote_code, num_labels=1
)
reward_model = AutoModelForSequenceClassification.from_pretrained(
training_args.reward_model_path, trust_remote_code=model_args.trust_remote_code, num_labels=1
)
policy = AutoModelForCausalLM.from_pretrained(
training_args.sft_model_path, trust_remote_code=model_args.trust_remote_code
)
peft_config = get_peft_config(model_args)
if peft_config is None:
ref_policy = AutoModelForCausalLM.from_pretrained(
training_args.sft_model_path, trust_remote_code=model_args.trust_remote_code
)
else:
ref_policy = None
################
# Dataset
################
dataset = load_dataset(
script_args.dataset_name, name=script_args.dataset_config, split=script_args.dataset_train_split
)
eval_samples = 100
train_dataset = dataset.select(range(len(dataset) - eval_samples))
eval_dataset = dataset.select(range(len(dataset) - eval_samples, len(dataset)))
dataset_text_field = "prompt"
def prepare_dataset(dataset, tokenizer):
"""pre-tokenize the dataset before training; only collate during training"""
def tokenize(element):
outputs = tokenizer(
element[dataset_text_field],
padding=False,
)
return {"input_ids": outputs["input_ids"]}
return dataset.map(
tokenize,
batched=True,
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)
################
# Training
################
trainer = PPOTrainer(
args=training_args,
processing_class=tokenizer,
model=policy,
ref_model=ref_policy,
reward_model=reward_model,
value_model=value_model,
train_dataset=train_dataset,
eval_dataset=eval_dataset,
peft_config=peft_config,
)
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()
|