Skip to content

add token ips #1161

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

Open
wants to merge 5 commits into
base: develop
Choose a base branch
from
Open

add token ips #1161

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f0507f6
add token ips
pkhk-1 Mar 26, 2025
f739171
add log
pkhk-1 Mar 26, 2025
42935f3
add some log
pkhk-1 Mar 26, 2025
f328318
fix log output
pkhk-1 Mar 26, 2025
9248193
Merge branch 'develop' into token_add
pkhk-1 Mar 26, 2025
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
23 changes: 2 additions & 21 deletions paddlemix/examples/llava/pretrain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,29 +163,10 @@ def main():
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
if training_args.benchmark:

def get_paddle_memory_info():
"""get_memory_info"""
divisor = 2**30
return (
paddle.device.cuda.memory_allocated() / divisor,
paddle.device.cuda.max_memory_allocated() / divisor,
paddle.device.cuda.memory_reserved() / divisor,
paddle.device.cuda.max_memory_reserved() / divisor,
)

memory_allocated, max_memory_allocated, memory_reserved, max_memory_reserved = get_paddle_memory_info()

logger.info(
f"memory_allocated:{memory_allocated}GB, max_memory_allocated: {max_memory_allocated}GB, memory_reserved:{memory_reserved}GB, max_memory_reserved: {max_memory_reserved}GB \n"
)
total_effective_samples = total_samples * training_args.num_train_epochs
effective_samples_per_second = total_effective_samples / train_result.metrics["train_runtime"]
mem_gpu = (
train_result.metrics["train_mem_gpu_peaked_delta"] + train_result.metrics["train_mem_gpu_alloc_delta"]
)
logger.info(f"ips: {effective_samples_per_second} ")
logger.info(f"train_mem_gpu_peaked: {int(mem_gpu/ (2**20))} MB")
logger.info(f"Effective_samples_per_second: {effective_samples_per_second} ")
logger.info(f"Train_runtime: {train_result.metrics['train_runtime']}")
logger.info("Benchmark done.")
else:
trainer.save_model(merge_tensor_parallel=training_args.tensor_parallel_degree > 1)
Expand Down
25 changes: 3 additions & 22 deletions paddlemix/examples/llava/supervised_finetune.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -188,30 +188,11 @@ def main():
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
if training_args.benchmark:

def get_paddle_memory_info():
"""get_memory_info"""
divisor = 2**30
return (
paddle.device.cuda.memory_allocated() / divisor,
paddle.device.cuda.max_memory_allocated() / divisor,
paddle.device.cuda.memory_reserved() / divisor,
paddle.device.cuda.max_memory_reserved() / divisor,
)

memory_allocated, max_memory_allocated, memory_reserved, max_memory_reserved = get_paddle_memory_info()

logger.info(
f"memory_allocated:{memory_allocated}GB, max_memory_allocated: {max_memory_allocated}GB, memory_reserved:{memory_reserved}GB, max_memory_reserved: {max_memory_reserved}GB \n"
)

total_effective_samples = total_samples * training_args.num_train_epochs
effective_samples_per_second = total_effective_samples / train_result.metrics["train_runtime"]
mem_gpu = (
train_result.metrics["train_mem_gpu_peaked_delta"] + train_result.metrics["train_mem_gpu_alloc_delta"]
)
logger.info(f"ips: {effective_samples_per_second} ")
logger.info(f"train_mem_gpu_peaked: {int(mem_gpu/ (2**20))} MB")

logger.info(f"Effective_samples_per_second: {effective_samples_per_second} ")
logger.info(f"Train_runtime: {train_result.metrics['train_runtime']}")
logger.info("Benchmark done.")
else:
trainer.save_model(merge_tensor_parallel=training_args.tensor_parallel_degree > 1)
Expand Down
10 changes: 10 additions & 0 deletions paddlemix/models/llava/language_model/llava_llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,6 +100,16 @@ def forward(
input_ids, position_ids, attention_mask, past_key_values, labels, images, image_size
)

# 通过attention_mask计算有效token数量
if attention_mask is not None:
# 统计当前batch的有效token数（排除padding）
current_batch_tokens = attention_mask.sum().item() # shape: (batch_size, seq_len)
else:
# 如果没有padding，直接取inputs_embeds的batch_size*seq_length
current_batch_tokens = inputs_embeds.size(0) * inputs_embeds.size(1)
self.efficient_token_count = current_batch_tokens
self.input_shape = inputs_embeds.shape

return super().forward(
input_ids=input_ids,
attention_mask=attention_mask,
Expand Down
Loading