Skip to content

[Optimization] Communication Optimization in TP Group #5596

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

Closed
ckl117 wants to merge 4 commits into from
+21 −8
Closed

[Optimization] Communication Optimization in TP Group #5596

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
326ab26
use cpu barrier sync tp_group, delete useless check and no sync token…
ckl117 Dec 16, 2025
edaa270
Merge branch 'develop' of https://github.com/PaddlePaddle/FastDeploy …
ckl117 Dec 16, 2025
a0d1d54
code check, add docs
ckl117 Dec 16, 2025
132635d
Merge branch 'develop' into dev_delete_tp_sync
ckl117 Dec 17, 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
3 changes: 3 additions & 0 deletions docs/usage/environment_variables.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,5 +88,8 @@ environment_variables: dict[str, Callable[[], Any]] = {

# Count for cache_transfer_manager process error
"FD_CACHE_PROC_ERROR_COUNT": lambda: int(os.getenv("FD_CACHE_PROC_ERROR_COUNT", "10")),

# Whether to force the inference engine to synchronize token_ids sampled by TP groups.
"FD_SYNC_TOKEN_IDS_ACROSS_TP": lambda: bool(int(os.getenv("FD_SYNC_TOKEN_IDS_ACROSS_TP", "0"))),
}
```
6 changes: 5 additions & 1 deletion docs/zh/usage/environment_variables.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,5 +87,9 @@ environment_variables: dict[str, Callable[[], Any]] = {
"FD_CACHE_PROC_EXIT_TIMEOUT": lambda: int(os.getenv("FD_CACHE_PROC_EXIT_TIMEOUT", "600")),

# cache_transfer_manager 进程残留时连续错误阈值
"FD_CACHE_PROC_ERROR_COUNT": lambda: int(os.getenv("FD_CACHE_PROC_ERROR_COUNT", "10")),}
"FD_CACHE_PROC_ERROR_COUNT": lambda: int(os.getenv("FD_CACHE_PROC_ERROR_COUNT", "10")),

# 是否强制推理引擎同步TP组采样到的 token_ids, 默认不同步
"FD_SYNC_TOKEN_IDS_ACROSS_TP": lambda: bool(int(os.getenv("FD_SYNC_TOKEN_IDS_ACROSS_TP", "0"))),
}
```
1 change: 1 addition & 0 deletions fastdeploy/envs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,6 +155,7 @@
"FD_OTLP_EXPORTER_SCHEDULE_DELAY_MILLIS": lambda: int(os.getenv("FD_OTLP_EXPORTER_SCHEDULE_DELAY_MILLIS", "500")),
"FD_OTLP_EXPORTER_MAX_EXPORT_BATCH_SIZE": lambda: int(os.getenv("FD_OTLP_EXPORTER_MAX_EXPORT_BATCH_SIZE", "64")),
"FD_TOKEN_PROCESSOR_HEALTH_TIMEOUT": lambda: int(os.getenv("FD_TOKEN_PROCESSOR_HEALTH_TIMEOUT", "120")),
"FD_SYNC_TOKEN_IDS_ACROSS_TP": lambda: bool(int(os.getenv("FD_SYNC_TOKEN_IDS_ACROSS_TP", "0"))),
}


Expand Down
10 changes: 8 additions & 2 deletions fastdeploy/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1854,7 +1854,10 @@ def _dummy_sampler_run(
self.share_inputs["stop_flags"],
)
sampler_output = self.sampler(logits, self.sampling_metadata)
if self.parallel_config.tensor_parallel_size > 1:
if (
envs.FD_SYNC_TOKEN_IDS_ACROSS_TP
or self.fd_config.structured_outputs_config.guided_decoding_backend != "off"
) and self.parallel_config.tensor_parallel_size > 1:
paddle.distributed.broadcast(
sampler_output.sampled_token_ids,
self.parallel_config.data_parallel_rank * self.parallel_config.tensor_parallel_size,
Expand Down Expand Up @@ -2447,7 +2450,10 @@ class at the server level, which is too granular for ModelRunner.
[sampler_output.sampled_token_ids.shape[0]], device="cpu", dtype="int64"
),
)
if self.parallel_config.tensor_parallel_size > 1:
if (
envs.FD_SYNC_TOKEN_IDS_ACROSS_TP
or self.fd_config.structured_outputs_config.guided_decoding_backend != "off"
) and self.parallel_config.tensor_parallel_size > 1:
paddle.distributed.broadcast(
sampler_output.sampled_token_ids,
self.parallel_config.data_parallel_rank * self.parallel_config.tensor_parallel_size,
Expand Down
9 changes: 4 additions & 5 deletions fastdeploy/worker/worker_process.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -289,10 +289,7 @@ def _broadcast_model_weights_signal(self, src: int, group) -> int:
return model_weights_signal_tensor.item()

def _tp_barrier_wait(self):
if current_platform.is_xpu():
self.task_queue.worker_process_tp_barrier.wait()
else:
paddle.distributed.barrier(self.parallel_config.tp_group)
self.task_queue.worker_process_tp_barrier.wait()

def _init_eplb_signal(self):
if not self.eplb_config.enable_eplb:
Expand Down Expand Up @@ -478,7 +475,9 @@ def event_loop_normal(self) -> None:
time.sleep(0.01)
continue

if self.exist_task_signal.value[0] == ExistTaskStatus.EXIST or self.task_queue.read_finish_flag.get() == 1:
if self.exist_task_signal.value[0] == ExistTaskStatus.EXIST or (
self.nnode > 1 and self.task_queue.read_finish_flag.get() == 1
):
logger.info(f"Rank: {self.local_rank} Detected new requests.")

tasks, read_finish = self.task_queue.get_tasks()
Expand Down
Loading