423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-17 04:10:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

update get template

Browse Source
This commit is contained in:
hiyouga
2024-09-04 22:36:20 +08:00
parent 8f441c2b3a
commit dabad5570b
17 changed files with 57 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llamafactory/train/dpo/workflow.py
View File

@@ -17,7 +17,7 @@
from typing import TYPE_CHECKING, List, Optional
from ...data import PairwiseDataCollatorWithPadding, get_dataset
from ...data import PairwiseDataCollatorWithPadding, get_dataset, get_template_and_fix_tokenizer
from ...extras.constants import IGNORE_INDEX
from ...extras.ploting import plot_loss
from ...hparams import ModelArguments
@@ -41,7 +41,8 @@ def run_dpo(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="rm", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="rm", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
data_collator = PairwiseDataCollatorWithPadding(

5
src/llamafactory/train/kto/workflow.py
View File

@@ -17,7 +17,7 @@
from typing import TYPE_CHECKING, List, Optional
from ...data import KTODataCollatorWithPadding, get_dataset
from ...data import KTODataCollatorWithPadding, get_dataset, get_template_and_fix_tokenizer
from ...extras.constants import IGNORE_INDEX
from ...extras.ploting import plot_loss
from ...hparams import ModelArguments
@@ -41,7 +41,8 @@ def run_kto(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="kto", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="kto", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
data_collator = KTODataCollatorWithPadding(

5
src/llamafactory/train/ppo/workflow.py
View File

@@ -17,7 +17,7 @@
from typing import TYPE_CHECKING, List, Optional
from ...data import MultiModalDataCollatorForSeq2Seq, get_dataset
from ...data import MultiModalDataCollatorForSeq2Seq, get_dataset, get_template_and_fix_tokenizer
from ...extras.ploting import plot_loss
from ...model import load_model, load_tokenizer
from ..callbacks import fix_valuehead_checkpoint
@@ -41,7 +41,8 @@ def run_ppo(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="ppo", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="ppo", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train, add_valuehead=True)
tokenizer.padding_side = "left" # use left-padding in generation while using right-padding in training

5
src/llamafactory/train/pt/workflow.py
View File

@@ -20,7 +20,7 @@ from typing import TYPE_CHECKING, List, Optional
from transformers import DataCollatorForLanguageModeling
from ...data import get_dataset
from ...data import get_dataset, get_template_and_fix_tokenizer
from ...extras.ploting import plot_loss
from ...model import load_model, load_tokenizer
from ..trainer_utils import create_modelcard_and_push
@@ -42,7 +42,8 @@ def run_pt(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, _ = get_dataset(model_args, data_args, training_args, stage="pt", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="pt", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
data_collator = DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False)

5
src/llamafactory/train/rm/workflow.py
View File

@@ -17,7 +17,7 @@
from typing import TYPE_CHECKING, List, Optional
from ...data import PairwiseDataCollatorWithPadding, get_dataset
from ...data import PairwiseDataCollatorWithPadding, get_dataset, get_template_and_fix_tokenizer
from ...extras.ploting import plot_loss
from ...model import load_model, load_tokenizer
from ..callbacks import fix_valuehead_checkpoint
@@ -41,7 +41,8 @@ def run_rm(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="rm", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="rm", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train, add_valuehead=True)
data_collator = PairwiseDataCollatorWithPadding(template=template, pad_to_multiple_of=8, **tokenizer_module)

7
src/llamafactory/train/sft/workflow.py
View File

@@ -17,7 +17,7 @@
from typing import TYPE_CHECKING, List, Optional
from ...data import SFTDataCollatorWith4DAttentionMask, get_dataset
from ...data import SFTDataCollatorWith4DAttentionMask, get_dataset, get_template_and_fix_tokenizer
from ...extras.constants import IGNORE_INDEX
from ...extras.misc import get_logits_processor
from ...extras.ploting import plot_loss
@@ -43,7 +43,8 @@ def run_sft(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="sft", **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer, data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, stage="sft", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
if getattr(model, "is_quantized", False) and not training_args.do_train:
@@ -62,7 +63,7 @@ def run_sft(
# 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 # important for multimodal and pairwise dataset
training_args.remove_unused_columns = False # important for multimodal dataset
# Metric utils
metric_module = {}

5
src/llamafactory/train/test_utils.py
View File

@@ -19,7 +19,7 @@ from peft import PeftModel
from transformers import AutoModelForCausalLM
from trl import AutoModelForCausalLMWithValueHead
from ..data import get_dataset
from ..data import get_dataset, get_template_and_fix_tokenizer
from ..extras.misc import get_current_device
from ..hparams import get_infer_args, get_train_args
from ..model import load_model, load_tokenizer
@@ -105,7 +105,8 @@ def load_reference_model(
def load_train_dataset(**kwargs) -> "Dataset":
model_args, data_args, training_args, _, _ = get_train_args(kwargs)
tokenizer_module = load_tokenizer(model_args)
dataset_module, _ = get_dataset(model_args, data_args, training_args, stage=kwargs["stage"], **tokenizer_module)
template = get_template_and_fix_tokenizer(tokenizer_module["tokenizer"], data_args)
dataset_module = get_dataset(template, model_args, data_args, training_args, kwargs["stage"], **tokenizer_module)
return dataset_module["train_dataset"]