Skip to content
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

Jump to bottom

refine disc pdll uts #1230

Merged
merged 1 commit into from
Jul 20, 2023
Merged

refine disc pdll uts #1230

Show file tree
Hide file tree
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
6 changes: 2 additions & 4 deletions pytorch_blade/tests/disc/pdl/test_e2e/test_quantization.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -395,7 +395,6 @@ def forward(self, x):
return x
model = Model().eval().to(self.device)
inp = torch.randn(512, 512).to(self.device)
traced_model = torch.jit.trace(model, inp)
# only cuda version 11.3/11.7 can be ensured correctness
if torch.version.cuda == '11.3':
qgemm_pdl_file = "dequant_gemm_quant_bias_quant.pdll"
Expand All @@ -406,7 +405,7 @@ def forward(self, x):
os.path.join(self.device_pdll_dir, qgemm_pdl_file)
]
pdll_files = ",".join(pdll_files)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, diff_scale=model.output_scale)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, atol=model.output_scale)

def test_s8s8s8_f32bias_per_tensor_three_rank_verify(self):
class Model(nn.Module):
Expand Down Expand Up @@ -448,7 +447,6 @@ def forward(self, x):
return x
model = Model().eval().to(self.device)
inp = torch.randn(8, 512, 512).to(self.device)
traced_model = torch.jit.trace(model, inp)
# only cuda version 11.3/11.7 can be ensured correctness
if torch.version.cuda == '11.3':
qgemm_pdl_file = "dequant_gemm_quant_bias_quant.pdll"
Expand All @@ -459,7 +457,7 @@ def forward(self, x):
os.path.join(self.device_pdll_dir, qgemm_pdl_file)
]
pdll_files = ",".join(pdll_files)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, diff_scale=model.output_scale)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, atol=model.output_scale)


if __name__ == "__main__":
Expand Down
5 changes: 3 additions & 2 deletions pytorch_blade/tests/disc/pdl/test_e2e/test_quantized_qkv_merge.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,8 @@ def forward(self, x):
x3, self.output_scale, self.output_zero_point,
self.activation_quant_min, self.activation_quant_max
)
return x1+x2+x3
return x1 + x2 + x3

model = Model().eval().to(self.device)
inp = torch.randn(512, 512).to(self.device)
traced_model = torch.jit.trace(model, inp)
Expand All @@ -106,7 +107,7 @@ def forward(self, x):
os.path.join(self.device_pdll_dir, qgemm_pdl_file)
]
pdll_files = ",".join(pdll_files)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, diff_scale=3*model.output_scale)
self._test_e2e(model, inp, pdll_files=pdll_files, enable_int8=True, atol=3 * model.output_scale)

if __name__ == "__main__":
unittest.main()
22 changes: 1 addition & 21 deletions pytorch_blade/tests/disc/testing_base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -18,7 +18,6 @@
from torch.testing import FileCheck
from torch_blade import mlir, optimize, utils
from torch_blade.clustering import support_fusion_group
from torch_blade.clustering.support_fusion_group import min_group_nodes
from torch_blade.config import Config
from torch_blade.mlir import is_available
from torch_blade.pass_manager import _optimize_common
Expand Down Expand Up @@ -153,7 +152,7 @@ def setUp(self):
def _test_torchscipte_to_mhlo(
self, module, expected_str, pdll_files=None,
pdll_dirs=None, enable_int8=False,
env_var = {},
env_var={},
):
if pdll_files is not None:
env_var["DISC_TORCH_PDL_FILES"] = pdll_files
Expand Down Expand Up @@ -225,22 +224,3 @@ def setUp(self):
super().setUp()
if self.device != torch.device('cuda'):
self.skipTest("Quantization pdl test case only supports gpu platform")

def _test_e2e(
self, model, inp, pdll_files=None,
pdll_dirs=None, enable_int8=False,
diff_scale=1.0
):
origin_output = model(inp)
cfg = Config.get_current_context_or_new()
cfg.optimization_pipeline = mlir.backend_name()
cfg.enable_int8 = enable_int8
env_var = {}
if pdll_files is not None:
env_var["DISC_TORCH_PDL_FILES"] = pdll_files
if pdll_dirs is not None:
env_var["DISC_TORCH_PDLL_INCLUDE_DIRS"] = pdll_dirs
with set_env(**env_var), cfg, min_group_nodes(1):
opt_model = optimize(model, True, inp)
now_output = opt_model(inp)
self.assertTrue(torch.allclose(now_output, origin_output, atol=1.0 * diff_scale))
Loading