Skip to content

[wip] add token log for llava bench #1065

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 1 commit into
base: develop
Choose a base branch
from
Open

[wip] add token log for llava bench #1065

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 8 additions & 1 deletion paddlemix/examples/llava/pretrain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,8 +171,15 @@ def get_paddle_memory_info():
mem_gpu = (
train_result.metrics["train_mem_gpu_peaked_delta"] + train_result.metrics["train_mem_gpu_alloc_delta"]
)
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')
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

callback的on_log方法中已经提供了这些性能数据，是否还需要重复log？


total_effective_samples = total_samples * training_args.num_train_epochs
effective_samples_per_second = total_effective_samples / train_result.metrics["train_runtime"]

logger.info(f"Effective_samples_per_second: {effective_samples_per_second} ")
logger.info(f"train_mem_gpu_peaked: {int(mem_gpu/ (2**20))} MB")
logger.info(f"avg_efficient_tokens_per_sec_per_card: {train_result.metrics['avg_efficient_tokens_per_sec_per_card']}")
logger.info(f"avg_tokens_per_sec_per_card: {train_result.metrics['avg_tokens_per_sec_per_card']}")

logger.info("Benchmark done.")
else:
trainer.save_model(merge_tensor_parallel=training_args.tensor_parallel_degree > 1)
Expand Down
11 changes: 11 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 @@ -99,6 +99,17 @@ def forward(
) = self.prepare_inputs_labels_for_multimodal(
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,
Expand Down
10 changes: 5 additions & 5 deletions paddlemix/tools/supervised_finetune.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,15 +191,15 @@ def get_paddle_memory_info():
)
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')
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"Effective_samples_per_second: {effective_samples_per_second} ")
logger.info(f"train_mem_gpu_peaked: {int(mem_gpu/ (2**20))} MB")
logger.info(f"avg_efficient_tokens_per_sec_per_card: {train_result.metrics['avg_efficient_tokens_per_sec_per_card']}")
logger.info(f"avg_tokens_per_sec_per_card: {train_result.metrics['avg_tokens_per_sec_per_card']}")

logger.info("Benchmark done.")
else:
trainer.save_model(merge_tensor_parallel=training_args.tensor_parallel_degree > 1)
Expand Down
Loading