github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-07-27 21:40:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

add profiling

Browse Source
This commit is contained in:
Tong Li 2025-06-26 17:49:53 +08:00
parent 71ef6b32c6
commit 58cb4fb4f7
2 changed files with 24 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
applications/ColossalChat/coati/distributed/consumer.py
View File

@ -134,8 +134,12 @@ class BaseConsumer:
def calculate_effective_group_to_raw_group_mapping(self, step): def calculate_effective_group_to_raw_group_mapping(self, step):
effective_group_to_raw_group_mapping = {} effective_group_to_raw_group_mapping = {}
for buffer_idx in range(len(self.buffer)): for buffer_idx in range(len(self.buffer)):
if self.buffer[buffer_idx][0] is not None and (self.buffer[buffer_idx][-1] <= step - self.n_behind): if self.buffer[buffer_idx][0] is not None:
effective_group_to_raw_group_mapping[len(effective_group_to_raw_group_mapping)] = buffer_idx if self.n_behind == 0:
effective_group_to_raw_group_mapping[len(effective_group_to_raw_group_mapping)] = buffer_idx
else:
if self.buffer[buffer_idx][-1] <= step - self.n_behind:
effective_group_to_raw_group_mapping[len(effective_group_to_raw_group_mapping)] = buffer_idx
return effective_group_to_raw_group_mapping return effective_group_to_raw_group_mapping
def loop(self) -> None: def loop(self) -> None:

25
applications/ColossalChat/rl_example.py
View File

@ -246,14 +246,25 @@ if __name__ == "__main__":
tensor_parallel_size=args.producer_tensor_parallel_size, tensor_parallel_size=args.producer_tensor_parallel_size,
) )
) )
generate_config.update( if args.enable_profiling:
dict( # If profiling is enabled, we force model to generate to max_new_tokens
max_tokens=args.max_new_tokens, # max new tokens inference_model_config.update(
ignore_eos=True if args.reward_type == "think_answer_tags" else False, dict(
include_stop_str_in_output=True, max_tokens=args.max_new_tokens, # max new tokens
stop=["</answer>"] if args.reward_type == "think_answer_tags" else None, ignore_eos=True,
include_stop_str_in_output=True,
stop=None,
)
)
else:
generate_config.update(
dict(
max_tokens=args.max_new_tokens, # max new tokens
ignore_eos=True if args.reward_type == "think_answer_tags" else False,
include_stop_str_in_output=True,
stop=["</answer>"] if args.reward_type == "think_answer_tags" else None,
)
) )
)
eval_generation_config = {"temperature": 0.6} # used to update generation config for evaluation eval_generation_config = {"temperature": 0.6} # used to update generation config for evaluation
else: else:
raise ValueError(f"Unsupported backend: {args.backend}") raise ValueError(f"Unsupported backend: {args.backend}")