ALinrunrun opened a new issue, #19559:
URL: https://github.com/apache/tvm/issues/19559
### Expected behavior
TVM Relax should execute ONNX `Sinh` and `Cosh` consistently with ONNX
Runtime for large but still representable float32 inputs.
For inputs around `x = 89`, ONNX Runtime returns finite float32 values close
to the upper range of float32.
### Actual behavior
TVM Relax returns `inf` / `-inf` while ONNX Runtime still returns finite
values:
```
Sinh input: [ 88.85 89. 89.2 -88.95]
ORT: [ 1.9321198e+38 2.2448064e+38 2.7418043e+38 -2.1353193e+38]
TVM: [ inf inf inf -inf]
Cosh input: [ 88.85 89. 89.2 -88.95]
ORT: [1.9321198e+38 2.2448064e+38 2.7418043e+38 2.1353193e+38]
TVM: [inf inf inf inf]
```
The discrepancy appears when importing ONNX Sinh and Cosh models through the
Relax ONNX frontend and compiling them for the llvm target.
### Environment
TVM: 0.14 environment / Relax ONNX frontend
ONNX Runtime: 1.23
Python: 3.11
Target: llvm
OS: Linux
### Steps to reproduce
```
import warnings
warnings.filterwarnings("ignore")
import numpy as np
import onnxruntime as ort
import tvm
from onnx import TensorProto, helper
from tvm import relax
from tvm.relax.frontend.onnx import from_onnx
def build_model(op):
node = helper.make_node(op, ["x"], ["y"])
graph = helper.make_graph(
[node],
"g",
[helper.make_tensor_value_info("x", TensorProto.FLOAT, [4])],
[helper.make_tensor_value_info("y", TensorProto.FLOAT, [4])],
)
model = helper.make_model(graph, opset_imports=[helper.make_opsetid("",
20)])
model.ir_version = 9
return model
x = np.array([88.85, 89.0, 89.2, -88.95], dtype=np.float32)
for op in ("Sinh", "Cosh"):
model = build_model(op)
sess = ort.InferenceSession(
model.SerializeToString(),
providers=["CPUExecutionProvider"],
)
ort_out = sess.run(None, {"x": x})[0]
mod = from_onnx(model)
with tvm.transform.PassContext(opt_level=3):
ex = tvm.compile(mod, target=tvm.target.Target("llvm"))
vm = relax.VirtualMachine(ex, tvm.cpu())
out = vm["main"](tvm.runtime.tensor(x, tvm.cpu()))
tvm_out = (out[0] if isinstance(out, (list, tuple)) else out).numpy()
print(f"{op} input:", x)
print(" ORT:", ort_out)
print(" TVM:", tvm_out)
```
### Triage
* needs-triage
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
