File size: 5,516 Bytes
2b915e2 |
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 |
# Copyright 2024 HuggingFace Inc. and the LlamaFactory team.
#
# This code is inspired by the HuggingFace's transformers library.
# https://github.com/huggingface/transformers/blob/v4.40.0/examples/pytorch/summarization/run_summarization.py
#
# 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.
from typing import TYPE_CHECKING, List, Optional
from ...data import SFTDataCollatorWith4DAttentionMask, get_dataset, split_dataset
from ...extras.constants import IGNORE_INDEX
from ...extras.misc import get_logits_processor
from ...extras.ploting import plot_loss
from ...model import load_model, load_tokenizer
from ..trainer_utils import create_modelcard_and_push
from .metric import ComputeMetrics, compute_accuracy, eval_logit_processor
from .trainer import CustomSeq2SeqTrainer
if TYPE_CHECKING:
from transformers import Seq2SeqTrainingArguments, TrainerCallback
from ...hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
def run_sft(
model_args: "ModelArguments",
data_args: "DataArguments",
training_args: "Seq2SeqTrainingArguments",
finetuning_args: "FinetuningArguments",
generating_args: "GeneratingArguments",
callbacks: Optional[List["TrainerCallback"]] = None,
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset = get_dataset(model_args, data_args, training_args, stage="sft", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
if training_args.predict_with_generate:
tokenizer.padding_side = "left" # use left-padding in generation
if getattr(model, "is_quantized", False) and not training_args.do_train:
setattr(model, "_hf_peft_config_loaded", True) # hack here: make model compatible with prediction
data_collator = SFTDataCollatorWith4DAttentionMask(
tokenizer=tokenizer,
pad_to_multiple_of=8 if tokenizer.padding_side == "right" else None, # for shift short attention
label_pad_token_id=IGNORE_INDEX if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id,
block_diag_attn=model_args.block_diag_attn,
attn_implementation=getattr(model.config, "_attn_implementation", None),
compute_dtype=model_args.compute_dtype,
)
# Override the decoding parameters of Seq2SeqTrainer
training_args.generation_max_length = training_args.generation_max_length or data_args.cutoff_len
training_args.generation_num_beams = data_args.eval_num_beams or training_args.generation_num_beams
training_args.remove_unused_columns = False if model_args.visual_inputs else training_args.remove_unused_columns
# Initialize our Trainer
trainer = CustomSeq2SeqTrainer(
model=model,
args=training_args,
finetuning_args=finetuning_args,
data_collator=data_collator,
callbacks=callbacks,
compute_metrics=ComputeMetrics(tokenizer) if training_args.predict_with_generate else compute_accuracy,
preprocess_logits_for_metrics=None if training_args.predict_with_generate else eval_logit_processor,
**tokenizer_module,
**split_dataset(dataset, data_args, training_args),
)
# Keyword arguments for `model.generate`
gen_kwargs = generating_args.to_dict()
gen_kwargs["eos_token_id"] = [tokenizer.eos_token_id] + tokenizer.additional_special_tokens_ids
gen_kwargs["pad_token_id"] = tokenizer.pad_token_id
gen_kwargs["logits_processor"] = get_logits_processor()
# Training
if training_args.do_train:
train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint)
trainer.save_model()
trainer.log_metrics("train", train_result.metrics)
trainer.save_metrics("train", train_result.metrics)
trainer.save_state()
if trainer.is_world_process_zero() and finetuning_args.plot_loss:
plot_loss(training_args.output_dir, keys=["loss", "eval_loss", "eval_accuracy"])
# Evaluation
if training_args.do_eval:
metrics = trainer.evaluate(metric_key_prefix="eval", **gen_kwargs)
if training_args.predict_with_generate: # eval_loss will be wrong if predict_with_generate is enabled
metrics.pop("eval_loss", None)
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)
# Predict
if training_args.do_predict:
predict_results = trainer.predict(dataset, metric_key_prefix="predict", **gen_kwargs)
if training_args.predict_with_generate: # predict_loss will be wrong if predict_with_generate is enabled
predict_results.metrics.pop("predict_loss", None)
trainer.log_metrics("predict", predict_results.metrics)
trainer.save_metrics("predict", predict_results.metrics)
trainer.save_predictions(dataset, predict_results)
# Create model card
create_modelcard_and_push(trainer, model_args, data_args, training_args, finetuning_args)
|