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

[Executor] Adjust signal sending order in RL training (#3773) (#4066) (#4178)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details

Browse Source 
* Adjust processing order

* fix bug

* fix update_parameters bug

* refine code
This commit is contained in:
RAM
2025-09-22 14:31:36 +08:00
committed by GitHub
parent 7bdc6f41e5
commit 01f6934162
3 changed files with 20 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
fastdeploy/worker/gpu_model_runner.py
View File

@@ -1705,25 +1705,27 @@ class GPUModelRunner(ModelRunnerBase):
paddle.device.cuda.empty_cache()
def clear_parameters(self, pid):
""" " Dynamic model loader use to clear parameters use for RL"""
"""Dynamic model loader use to clear parameters use for RL"""
# Clear CUDAGraph
if self.use_cudagraph:
self.model.clear_grpah_opt_backend()
# Clear parameters and Send single
self.dynamic_weight_manager.clear_parameters(pid)
self.clear_cache()
paddle.device.cuda.empty_cache()
# Clear CudaGraph
if self.use_cudagraph:
self.model.clear_grpah_opt_backend()
self.dynamic_weight_manager._log_memory("dynamic weight manager clear all memory")
def update_parameters(self, pid):
""" " Dynamic model loader use to update parameters use for RL"""
"""Dynamic model loader use to update parameters use for RL"""
# Update parameters
self.dynamic_weight_manager.update_parameters(pid)
self.initialize_kv_cache()
# Recapture CudaGraph
# Recapture CUDAGraph
if self.use_cudagraph:
self.capture_model()
# Send single
self.dynamic_weight_manager.finalize_update(pid)
self.dynamic_weight_manager._log_memory("dynamic weight manager update all memory")