Skip to content

[just test] tmp fix k3 in step0 #7897

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
zoooo0820 wants to merge 1 commit into
base: develop
Choose a base branch
from
+6 −6
Open

[just test] tmp fix k3 in step0 #7897

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
12 changes: 6 additions & 6 deletions fastdeploy/worker/gpu_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -242,12 +242,6 @@ def graph_optimize_and_warm_up_model(self) -> None:
):
self.model_runner.capture_model_prefill_and_mixed()

# Capture CUDAGraph for decode phase (all modes)
self.model_runner.capture_model()

# Block-wise CUDA graph capture (independent loop)
self.model_runner.capture_block_wise_graphs()

# Deterministic mode: reset RNG and share_inputs after warmup.
# Warmup _dummy_run() calls consume CUDA RNG state and leave stale
# data (infer_seed, stop_flags, seq_lens, etc.) in share_inputs.
Expand All @@ -257,6 +251,12 @@ def graph_optimize_and_warm_up_model(self) -> None:
set_random_seed(self.fd_config.model_config.seed)
self.model_runner.share_inputs.reset_share_inputs()

# Capture CUDAGraph for decode phase (all modes)
self.model_runner.capture_model()
Copy link
Copy Markdown

@PaddlePaddle-bot PaddlePaddle-bot May 22, 2026

Choose a reason for hiding this comment

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

🟡 建议 多硬件同步检查

gpu_worker.py 修改了 CUDAGraph capture 的执行顺序（capture_model() 移到 deterministic reset 之后），如果其他硬件 Worker（dcu_worker.py、xpu_worker.py、iluvatar_worker.py、metax_worker.py 等）中有相同的 CUDAGraph capture 逻辑，是否也存在同样的 deterministic 模式问题？

建议：检查 fastdeploy/worker/{xpu,dcu,gcu,hpu,iluvatar,metax}_worker.py 中是否有类似 capture 顺序问题，如有，同步修复。


# Block-wise CUDA graph capture (independent loop)
self.model_runner.capture_block_wise_graphs()

def check_health(self) -> bool:
""" """
return True
Expand Down
Loading