saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-30 20:18:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Partially disable QGemm tests for float 8 types (#18196)

Browse source 
### Description
The quantization tool assumes QGemm is implemented for float 8 types but
it is not yet supported. The condition partially disabling the test was
not robust enough. This is changed by this PR.
This commit is contained in:
Xavier Dupré 2023-11-03 10:17:50 +01:00 committed by GitHub
parent 9f5a6856fe
commit 1439da36fe
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

21
onnxruntime/test/python/quantization/test_op_gemm.py
View file

@ -192,24 +192,9 @@ class TestOpGemm(unittest.TestCase):
check_qtype_by_node_type(self, model_int8_path, qnode_io_qtypes)
data_reader.rewind()
if activation_type_str == "f8e4m3fn" and weight_type_str == "f8e4m3fn":
# QGemm is not implemented for CPU.
try:
check_model_correctness(
self,
model_fp32_path,
model_int8_path,
data_reader.get_next(),
providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
is_gemm=True,
)
except Exception as e:
if (
"Type 'tensor(float8e4m3fn)' of input parameter (input_quantized) of operator (QGemm) in node () is invalid."
in str(e)
):
warnings.warn("Fix this test when QGemm is implemented.")
return
raise e
# QGemm for float 8 is not implemented. The test should be updated when it is.
warnings.warn("Fix this test when QGemm is implemented for float 8 types.")
return
else:
check_model_correctness(self, model_fp32_path, model_int8_path, data_reader.get_next(), is_gemm=True)