apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix bug for pd step signal (#3230)

Browse Source
This commit is contained in:
chenjian
2025-08-06 10:41:52 +08:00
committed by GitHub
parent a9d231c900
commit a4572a5e5d
2 changed files with 6 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
fastdeploy/splitwise/internal_adapter_utils.py
View File

@@ -61,6 +61,7 @@ class InternalAdapter:
"max_batch_size": int(available_batch_size), "max_batch_size": int(available_batch_size),
"max_input_token_num": self.cfg.max_num_batched_tokens, "max_input_token_num": self.cfg.max_num_batched_tokens,
"unhandled_request_num": self.engine.scheduler.get_unhandled_request_num(), "unhandled_request_num": self.engine.scheduler.get_unhandled_request_num(),
"available_batch": int(self.engine.resource_manager.available_batch()),
} }
return server_info return server_info

10
fastdeploy/worker/gpu_model_runner.py
View File

@@ -1268,11 +1268,6 @@ class GPUModelRunner(ModelRunnerBase):
We plan to replace it with 'ModelForwardBatch'. We plan to replace it with 'ModelForwardBatch'.
intermediate_tensors: intermediate_tensors:
""" """
# 1. Prepare inputs of model and sampler.
skip_idx_list = self._get_skip_idx(model_forward_batch)
self._prepare_inputs()
self.sampler.pre_process(skip_idx_list)
# NOTE(wufeisheng): If `not_need_stop`` is False, it means the current worker is in an idle state. # NOTE(wufeisheng): If `not_need_stop`` is False, it means the current worker is in an idle state.
# This logic is not used in TP (Tensor Parallelism) mode. However, in EP (Expert Parallelism) mode, # This logic is not used in TP (Tensor Parallelism) mode. However, in EP (Expert Parallelism) mode,
# when there is data on other runner, the current runner is required to execute part of the model. # when there is data on other runner, the current runner is required to execute part of the model.
@@ -1280,6 +1275,11 @@ class GPUModelRunner(ModelRunnerBase):
self._execute_empty_input() self._execute_empty_input()
return None return None
# 1. Prepare inputs of model and sampler.
skip_idx_list = self._get_skip_idx(model_forward_batch)
self._prepare_inputs()
self.sampler.pre_process(skip_idx_list)
# 2. Padding inputs for cuda graph # 2. Padding inputs for cuda graph
self.padding_cudagraph_inputs() self.padding_cudagraph_inputs()