Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Feature] Support the DatasetInfoHook of DPO training #787

Merged
merged 2 commits into from
Jul 11, 2024
Merged

[Feature] Support the DatasetInfoHook of DPO training #787

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e999025
[Feature] Support the DatasetInfoHook of DPO training
xu-song Jun 20, 2024
a222889
fix yapf check
xu-song Jul 9, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions xtuner/configs/dpo/internlm/internlm2_chat_1_8b_dpo_full.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
preference_collate_fn
from xtuner.dataset.preference_dataset import (build_preference_dataset,
orpo_dpo_mix_40k_map_fn)
from xtuner.engine.hooks import (EvaluateChatHook,
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model.dpo import DPO
Expand Down Expand Up @@ -141,7 +141,7 @@
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
# dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
Expand Down
4 changes: 2 additions & 2 deletions xtuner/configs/dpo/internlm/internlm2_chat_1_8b_dpo_full_varlenattn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
preference_collate_fn
from xtuner.dataset.preference_dataset import (build_preference_dataset,
orpo_dpo_mix_40k_map_fn)
from xtuner.engine.hooks import (EvaluateChatHook,
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model.dpo import DPO
Expand Down Expand Up @@ -151,7 +151,7 @@
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
# dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
Expand Down
4 changes: 2 additions & 2 deletions xtuner/configs/dpo/internlm/internlm2_chat_1_8b_dpo_full_varlenattn_jsonl_dataset.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@
preference_collate_fn
from xtuner.dataset.preference_dataset import (build_preference_dataset,
load_jsonl_dataset)
from xtuner.engine.hooks import (EvaluateChatHook,
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model.dpo import DPO
Expand Down Expand Up @@ -155,7 +155,7 @@
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
# dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
Expand Down
4 changes: 2 additions & 2 deletions xtuner/configs/dpo/internlm/internlm2_chat_7b_dpo_qlora_varlenattn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@
preference_collate_fn
from xtuner.dataset.preference_dataset import (build_preference_dataset,
orpo_dpo_mix_40k_map_fn)
from xtuner.engine.hooks import (EvaluateChatHook,
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model.dpo import DPO
Expand Down Expand Up @@ -170,7 +170,7 @@
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
# dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
Expand Down
4 changes: 2 additions & 2 deletions xtuner/configs/dpo/llama/llama3_8b_instruct_dpo_qlora_varlenattn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@
preference_collate_fn
from xtuner.dataset.preference_dataset import (build_preference_dataset,
orpo_dpo_mix_40k_map_fn)
from xtuner.engine.hooks import (EvaluateChatHook,
from xtuner.engine.hooks import (DatasetInfoHook, EvaluateChatHook,
VarlenAttnArgsToMessageHubHook)
from xtuner.engine.runner import TrainLoop
from xtuner.model.dpo import DPO
Expand Down Expand Up @@ -170,7 +170,7 @@
#######################################################################
# Log the dialogue periodically during the training process, optional
custom_hooks = [
# dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(type=DatasetInfoHook, tokenizer=tokenizer),
dict(
type=EvaluateChatHook,
tokenizer=tokenizer,
Expand Down
29 changes: 18 additions & 11 deletions xtuner/engine/hooks/dataset_info_hook.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,19 +25,26 @@ def __init__(self, tokenizer, is_intern_repo_dataset=False):
self.is_intern_repo_dataset = is_intern_repo_dataset

def log(self, runner, dataset, mode='train'):

def _log(input_ids, log_prefix=''):
if self.is_intern_repo_dataset:
input_ids = [abs(x) for x in input_ids]
# Try to split list to be compatible with IMAGE token
input_ids = split_list(input_ids, IMAGE_TOKEN_INDEX)
text = log_prefix
for idx, ids in enumerate(input_ids):
text += self.tokenizer.decode(ids)
if idx != len(input_ids) - 1:
text += DEFAULT_IMAGE_TOKEN
runner.logger.info(text)

runner.logger.info(f'Num {mode} samples {len(dataset)}')
runner.logger.info(f'{mode} example:')
input_ids = dataset[0]['input_ids']
if self.is_intern_repo_dataset:
input_ids = [abs(x) for x in input_ids]
# Try to split list to be compatible with IMAGE token
input_ids = split_list(input_ids, IMAGE_TOKEN_INDEX)
text = ''
for idx, ids in enumerate(input_ids):
text += self.tokenizer.decode(ids)
if idx != len(input_ids) - 1:
text += DEFAULT_IMAGE_TOKEN
runner.logger.info(text)
if 'chosen_ids' in dataset[0]:
_log(dataset[0]['chosen_ids'], log_prefix='chosen: ')
_log(dataset[0]['rejected_ids'], log_prefix='rejected: ')
else:
_log(dataset[0]['input_ids'])

def before_train(self, runner) -> None:
do_train = runner.train_loop is not None
Expand Down
Loading