[SOT] Change warnings to errors and remove fallback operations (#4378)

* Change warnings to errors and remove fallback operations

* fix unitest

* fix codestyle
This commit is contained in:
Ryan
2025-10-17 11:27:04 +08:00
committed by GitHub
parent 0413c32b8f
commit 6160145f82
3 changed files with 27 additions and 28 deletions
@@ -14,17 +14,17 @@
# limitations under the License.
"""
import os
os.environ["FLAGS_cuda_graph_blacklist"] = "pd_op.matmul,pd_op.transpose"
import unittest
from unittest.mock import Mock
import paddle
import paddle.nn as nn
from fastdeploy.model_executor.graph_optimization.utils import sot_warmup_guard
paddle.set_flags({"FLAGS_cuda_graph_blacklist": "pd_op.matmul,pd_op.transpose"})
from fastdeploy.config import (
CacheConfig,
FDConfig,
@@ -77,10 +77,10 @@ class TestModel(nn.Layer):
super().__init__()
self.model = Attention(fd_config)
def forward(self, ids_remove_padding, forward_meta: ForwardMeta):
def forward(self, ids_remove_padding: paddle.Tensor, forward_meta: ForwardMeta):
return self.model(ids_remove_padding=ids_remove_padding, forward_meta=forward_meta)
def forward_correct(self, ids_remove_padding, forward_meta: ForwardMeta):
def forward_correct(self, ids_remove_padding: paddle.Tensor, forward_meta: ForwardMeta):
return self.model.forward_dynamic(ids_remove_padding=ids_remove_padding, forward_meta=forward_meta)
@@ -111,7 +111,8 @@ class TestStaticGraphCUDAGraphSplit(unittest.TestCase):
forward_meta1 = ForwardMeta(input_ids=x, ids_remove_padding=x, step_use_cudagraph=True)
# Trigger Capture
_ = test_model1(x, forward_meta=forward_meta1)
with sot_warmup_guard(True):
_ = test_model1(x, forward_meta=forward_meta1)
# Replay
_ = test_model1(x, forward_meta=forward_meta1)