ChatGLM #527
Merged
ChatGLM #527
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
xiezipeng-ML
reviewed
Jan 2, 2024
xiezipeng-ML
reviewed
Jan 2, 2024
xiezipeng-ML
reviewed
Jan 2, 2024
eval 结果{'results': {'hellaswag': {'acc': 0.4439354710217088, 'acc_stderr': 0.0049583141142664905, 'acc_norm': 0.5696076478789086, 'acc_norm_stderr': 0.0049411916073179105}}, 'versions': {'hellaswag': 0}, 'config': {'model': 'chatglm', 'batch_size': 1, 'device': 'cuda:0', 'num_fewshot': 0, 'limit': None, 'bootstrap_iters': 100000}} 尚存在问题训练阶段最后会卡住(trainer.train()结束后会卡住不动)，可以正常训练和保存所有模型 |
|
该Pr完成了: chatglm-6B的微调（lora微调/全量），推理; [lora部分也可移植到其他模型] 显存占用情况1. full finetune1n4g[1-4] fp16 1dp 4tp 1pp batch_size=1 [01/09 18:34:36 lb.utils.events]: eta: 17:02:20 iteration: 9/27736 consumed_samples: 80 total_loss: 5.674 time: 2.2448 s/iter data_time: 0.1167 s/iter total_throughput: 3.56 samples/s lr: 1.62e-08 1n4g[1-4] fp16 1dp 1tp 4pp batch_size=1 [01/09 18:43:45 lb.utils.events]: eta: 10:05:12 iteration: 9/27736 consumed_samples: 80 total_loss: 5.674 time: 1.3446 s/iter data_time: 0.0538 s/iter total_throughput: 5.95 samples/s lr: 1.62e-08 2. lora finetune1n4g[1-4] fp16 1dp 4tp 1pp batch_size=1 [01/09 18:55:21 lb.utils.events]: eta: 12:51:07 iteration: 9/27736 consumed_samples: 80 total_loss: 5.674 time: 1.7432 s/iter data_time: 0.0278 s/iter total_throughput: 4.59 samples/s lr: 1.62e-08 1n4g[1-4] fp16 1dp 1tp 4pp batch_size=1 [01/09 19:01:28 lb.utils.events]: eta: 6:29:57 iteration: 9/27736 consumed_samples: 80 total_loss: 5.674 time: 0.8229 s/iter data_time: 0.0110 s/iter total_throughput: 9.72 samples/s lr: 1.62e-08 |
def main(args):
cfg = LazyConfig.load(args.config_file)
cfg = LazyConfig.apply_overrides(cfg, args.opts)
default_setup(cfg, args)
seed_for_rank = cfg.train.seed + flow.env.get_rank()
flow.manual_seed(seed_for_rank)
flow.cuda.manual_seed(seed_for_rank)
np.random.seed(seed_for_rank)
random.seed(seed_for_rank)
if args.fast_dev_run:
cfg.train.train_epoch = 0
cfg.train.train_iter = 20
cfg.train.evaluation.eval_period = 10
cfg.train.log_period = 1
if args.eval_only:
tokenizer = None
if try_get_key(cfg, "tokenization") is not None:
tokenizer = DefaultTrainer.build_tokenizer(cfg)
model = DefaultTrainer.build_model(cfg)
Checkpointer(model, save_dir=cfg.train.output_dir).resume_or_load(
cfg.train.load_weight, resume=args.resume
)
if try_get_key(cfg, "graph.enabled", default=False):
model = DefaultTrainer.build_graph(cfg, model, is_train=False)
test_loader = DefaultTrainer.build_test_loader(cfg, tokenizer)
if len(test_loader) == 0:
logger = logging.getLogger(__name__)
logger.info("No dataset in dataloader.test, please set dataset for dataloader.test")
_ = DefaultTrainer.test(cfg, test_loader, model)
return
trainer = ChatGLMTrainer(cfg)
return trainer.train()
if __name__ == "__main__":
args = default_argument_parser().parse_args()
main(args)在训练结束后(保存完所有模型，并且在libai中trainer.train()完成后)，会卡住不动一段时间，然后发生下面的报错 版本信息 |
loxs123
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
loxs123
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
loxs123
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
loxs123
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
loxs123
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
xiezipeng-ML
approved these changes
Jan 12, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.