FP8 triton-ascend probe: /home/l30073763/fp8/fp8_triton_ascend_prob.py
Python: /root/miniconda3/envs/ly_inductor_2_2/bin/python
Timeout per probe: 180s
Selected probes: environment, tensor_e4m3, tensor_e5m2, copy_e4m3, copy_e5m2, cast_roundtrip_e4m3, cast_roundtrip_e5m2, pointwise_e4m3, pointwise_e5m2, amax_e4m3, amax_e5m2, dot_e4m3_e4m3, dot_e5m2_e5m2, dot_e4m3_e5m2, dot_scaled_e4m3, dot_scaled_e5m2, torch_scaled_mm_e4m3

[1/17] RUN environment
FP8_PROBE_CHILD_RESULT={"details": {"current_device": 0, "device_name": "Ascend950PR_9599", "device_properties": "_NPUDeviceProperties(name='Ascend950PR_9599', total_memory=126110MB, cube_core_num=36, vector_core_num=72, uuid=0008d80a-d200-0000-0000-000000009200, L2_cache_size=128MB)", "env": {"ASCEND_LAUNCH_BLOCKING": "1", "NPU_IS_A5": null, "TORCHINDUCTOR_NPU_BACKEND": null, "TRITON_ASCEND_ARCH": null}, "npu_available": true, "npu_count": 1, "package_torch-npu": "2.13.0", "package_triton": "3.5.0", "package_triton-ascend": "3.2.2+dev20260727051020", "platform": "Linux-6.6.0-159.4.3.154.oe2403sp4.aarch64-aarch64-with-glibc2.34", "python": "3.10.0 (default, Mar 3 2022, 09:51:40) [GCC 10.2.0]", "torch_npu_soc_version": 260, "torch_npu_version": "2.13.0", "torch_version": "2.13.0+cpu", "triton_compile_on_910_95": true, "triton_is_compile_on_910_95": true, "triton_supported_fp8_dtypes": ["fp8e5", "fp8e4b15", "fp8e4nv", "fp8e4b8", "fp8e5b16"], "triton_version_attr": "3.2.0"}, "duration_seconds": 0.379, "name": "environment", "status": "PASS"}
[1/17] PASS environment

[2/17] RUN tensor_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"dtype": "torch.float8_e4m3fn", "element_size": 1, "finite": true, "max": 1.5, "min": -1.5, "numel": 257}, "duration_seconds": 2.691, "name": "tensor_e4m3", "status": "PASS"}
[2/17] PASS tensor_e4m3

[3/17] RUN tensor_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"dtype": "torch.float8_e5m2", "element_size": 1, "finite": true, "max": 1.5, "min": -1.5, "numel": 257}, "duration_seconds": 2.85, "name": "tensor_e5m2", "status": "PASS"}
[3/17] PASS tensor_e5m2

[4/17] RUN copy_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"bitwise_equal": true, "comparison": "uint8_view", "numel": 257}, "duration_seconds": 2.798, "name": "copy_e4m3", "status": "PASS"}
[4/17] PASS copy_e4m3

[5/17] RUN copy_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"bitwise_equal": true, "comparison": "uint8_view", "numel": 257}, "duration_seconds": 2.797, "name": "copy_e5m2", "status": "PASS"}
[5/17] PASS copy_e5m2

[6/17] RUN cast_roundtrip_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"fp32_to_fp8_bitwise_equal": true, "fp32_to_fp8_comparison": "uint8_view", "fp32_to_fp8_numeric": {"allclose": true, "atol": 0.0, "dtype": "torch.float8_e4m3fn", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": [257]}, "fp8_to_fp32": {"allclose": true, "atol": 0.0, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": [257]}}, "duration_seconds": 2.875, "name": "cast_roundtrip_e4m3", "status": "PASS"}
[6/17] PASS cast_roundtrip_e4m3

[7/17] RUN cast_roundtrip_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"fp32_to_fp8_bitwise_equal": true, "fp32_to_fp8_comparison": "uint8_view", "fp32_to_fp8_numeric": {"allclose": true, "atol": 0.0, "dtype": "torch.float8_e5m2", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": [257]}, "fp8_to_fp32": {"allclose": true, "atol": 0.0, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": [257]}}, "duration_seconds": 2.82, "name": "cast_roundtrip_e5m2", "status": "PASS"}
[7/17] PASS cast_roundtrip_e5m2

[8/17] RUN pointwise_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 1e-05, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 1e-05, "shape": [257]}, "duration_seconds": 2.848, "name": "pointwise_e4m3", "status": "PASS"}
[8/17] PASS pointwise_e4m3

[9/17] RUN pointwise_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 1e-05, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 1e-05, "shape": [257]}, "duration_seconds": 2.846, "name": "pointwise_e5m2", "status": "PASS"}
[9/17] PASS pointwise_e5m2

[10/17] RUN amax_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 0.0, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": []}, "duration_seconds": 2.848, "name": "amax_e4m3", "status": "PASS"}
[10/17] PASS amax_e4m3

[11/17] RUN amax_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 0.0, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.0, "shape": []}, "duration_seconds": 2.837, "name": "amax_e5m2", "status": "PASS"}
[11/17] PASS amax_e5m2

[12/17] RUN dot_e4m3_e4m3
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.1, "shape": [16, 16]}, "duration_seconds": 2.902, "name": "dot_e4m3_e4m3", "status": "PASS"}
[12/17] PASS dot_e4m3_e4m3

[13/17] RUN dot_e5m2_e5m2
FP8_PROBE_CHILD_RESULT={"details": {"allclose": true, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.0, "mean_abs_error": 0.0, "rtol": 0.1, "shape": [16, 16]}, "duration_seconds": 2.862, "name": "dot_e5m2_e5m2", "status": "PASS"}
[13/17] PASS dot_e5m2_e5m2

[14/17] RUN dot_e4m3_e5m2
FP8_PROBE_CHILD_RESULT={"duration_seconds": 2.892, "error": "{"allclose": false, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.64825439453125, "mean_abs_error": 0.25658777356147766, "rtol": 0.1, "shape": [16, 16]}", "error_type": "AssertionError", "name": "dot_e4m3_e5m2", "status": "FAIL", "traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 499, in _run_named_probe\n return _probe_dot("e4m3", "e5m2")\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 425, in _probe_dot\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.64825439453125, "mean_abs_error": 0.25658777356147766, "rtol": 0.1, "shape": [16, 16]}\n"}
[14/17] FAIL dot_e4m3_e5m2

[15/17] RUN dot_scaled_e4m3
FP8_PROBE_CHILD_RESULT={"duration_seconds": 2.881, "error": "{"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1947364807128906, "mean_abs_error": 0.5505611300468445, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}", "error_type": "AssertionError", "name": "dot_scaled_e4m3", "status": "FAIL", "traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 501, in _run_named_probe\n return probe_dot_scaled(name.removeprefix("dot_scaled"))\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 458, in _probe_dot_scaled\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1947364807128906, "mean_abs_error": 0.5505611300468445, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}\n"}
[15/17] FAIL dot_scaled_e4m3

[16/17] RUN dot_scaled_e5m2
FP8_PROBE_CHILD_RESULT={"duration_seconds": 2.857, "error": "{"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1368446350097656, "mean_abs_error": 0.5454071760177612, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}", "error_type": "AssertionError", "name": "dot_scaled_e5m2", "status": "FAIL", "traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 501, in _run_named_probe\n return probe_dot_scaled(name.removeprefix("dot_scaled"))\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 458, in _probe_dot_scaled\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1368446350097656, "mean_abs_error": 0.5454071760177612, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}\n"}
[16/17] FAIL dot_scaled_e5m2

[17/17] RUN torch_scaled_mm_e4m3
FP8_PROBE_CHILD_RESULT={"duration_seconds": 2.688, "error": "_scaled_mm:../third_party/op-plugin/op_plugin/ops/opapi/ScaledMmKernelNpuOpApi.cpp:433 NPU function error: call aclnnQuantMatmulV5 failed, error code is 161002\n[ERROR] 2026-07-31-08:55:21 (PID:1494979, Device:0, RankID:-1) ERR00100 PTA call acl api failed.\n[PID: 1494979] 2026-07-31-08:55:21.388.591 Invalid_Argument_Tensor_Dtype(EZ0020): Parameter out of aclnnQuantMatmulV5GetWorkspaceSize has incorrect dtype DT_INT8. Reason: the dtype of out must be in dtype support list [DT_FLOAT16,DT_BFLOAT16,DT_FLOAT,].\n Solution: Check whether the dtype of input/output tensor meets the condition.\n", "error_type": "RuntimeError", "name": "torch_scaled_mm_e4m3", "status": "FAIL", "traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 503, in _run_named_probe\n return _probe_torch_scaled_mm()\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 469, in _probe_torch_scaled_mm\n out = torch._scaled_mm(a, b, scale_a, scale_b)\nRuntimeError: _scaled_mm:../third_party/op-plugin/op_plugin/ops/opapi/ScaledMmKernelNpuOpApi.cpp:433 NPU function error: call aclnnQuantMatmulV5 failed, error code is 161002\n[ERROR] 2026-07-31-08:55:21 (PID:1494979, Device:0, RankID:-1) ERR00100 PTA call acl api failed.\n[PID: 1494979] 2026-07-31-08:55:21.388.591 Invalid_Argument_Tensor_Dtype(EZ0020): Parameter out of aclnnQuantMatmulV5GetWorkspaceSize has incorrect dtype DT_INT8. Reason: the dtype of out must be in dtype support list [DT_FLOAT16,DT_BFLOAT16,DT_FLOAT,].\n Solution: Check whether the dtype of input/output tensor meets the condition.\n\n"}
[17/17] FAIL torch_scaled_mm_e4m3

FP8_PROBE_SUMMARY_BEGIN
{
"counts": {
"FAIL": 4,
"PASS": 13
},
"python": "/root/miniconda3/envs/ly_inductor_2_2/bin/python",
"results": [
{
"details": {
"current_device": 0,
"device_name": "Ascend950PR_9599",
"device_properties": "_NPUDeviceProperties(name='Ascend950PR_9599', total_memory=126110MB, cube_core_num=36, vector_core_num=72, uuid=0008d80a-d200-0000-0000-000000009200, L2_cache_size=128MB)",
"env": {
"ASCEND_LAUNCH_BLOCKING": "1",
"NPU_IS_A5": null,
"TORCHINDUCTOR_NPU_BACKEND": null,
"TRITON_ASCEND_ARCH": null
},
"npu_available": true,
"npu_count": 1,
"package_torch-npu": "2.13.0",
"package_triton": "3.5.0",
"package_triton-ascend": "3.2.2+dev20260727051020",
"platform": "Linux-6.6.0-159.4.3.154.oe2403sp4.aarch64-aarch64-with-glibc2.34",
"python": "3.10.0 (default, Mar 3 2022, 09:51:40) [GCC 10.2.0]",
"torch_npu_soc_version": 260,
"torch_npu_version": "2.13.0",
"torch_version": "2.13.0+cpu",
"triton_compile_on_910_95": true,
"triton_is_compile_on_910_95": true,
"triton_supported_fp8_dtypes": [
"fp8e5",
"fp8e4b15",
"fp8e4nv",
"fp8e4b8",
"fp8e5b16"
],
"triton_version_attr": "3.2.0"
},
"duration_seconds": 0.379,
"name": "environment",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"dtype": "torch.float8_e4m3fn",
"element_size": 1,
"finite": true,
"max": 1.5,
"min": -1.5,
"numel": 257
},
"duration_seconds": 2.691,
"name": "tensor_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"dtype": "torch.float8_e5m2",
"element_size": 1,
"finite": true,
"max": 1.5,
"min": -1.5,
"numel": 257
},
"duration_seconds": 2.85,
"name": "tensor_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"bitwise_equal": true,
"comparison": "uint8_view",
"numel": 257
},
"duration_seconds": 2.798,
"name": "copy_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"bitwise_equal": true,
"comparison": "uint8_view",
"numel": 257
},
"duration_seconds": 2.797,
"name": "copy_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"fp32_to_fp8_bitwise_equal": true,
"fp32_to_fp8_comparison": "uint8_view",
"fp32_to_fp8_numeric": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float8_e4m3fn",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": [
257
]
},
"fp8_to_fp32": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": [
257
]
}
},
"duration_seconds": 2.875,
"name": "cast_roundtrip_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"fp32_to_fp8_bitwise_equal": true,
"fp32_to_fp8_comparison": "uint8_view",
"fp32_to_fp8_numeric": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float8_e5m2",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": [
257
]
},
"fp8_to_fp32": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": [
257
]
}
},
"duration_seconds": 2.82,
"name": "cast_roundtrip_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 1e-05,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 1e-05,
"shape": [
257
]
},
"duration_seconds": 2.848,
"name": "pointwise_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 1e-05,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 1e-05,
"shape": [
257
]
},
"duration_seconds": 2.846,
"name": "pointwise_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": []
},
"duration_seconds": 2.848,
"name": "amax_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 0.0,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.0,
"shape": []
},
"duration_seconds": 2.837,
"name": "amax_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 0.25,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.1,
"shape": [
16,
16
]
},
"duration_seconds": 2.902,
"name": "dot_e4m3_e4m3",
"returncode": 0,
"status": "PASS"
},
{
"details": {
"allclose": true,
"atol": 0.25,
"dtype": "torch.float32",
"finite": true,
"max_abs_error": 0.0,
"mean_abs_error": 0.0,
"rtol": 0.1,
"shape": [
16,
16
]
},
"duration_seconds": 2.862,
"name": "dot_e5m2_e5m2",
"returncode": 0,
"status": "PASS"
},
{
"duration_seconds": 2.892,
"error": "{"allclose": false, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.64825439453125, "mean_abs_error": 0.25658777356147766, "rtol": 0.1, "shape": [16, 16]}",
"error_type": "AssertionError",
"name": "dot_e4m3_e5m2",
"returncode": 1,
"status": "FAIL",
"traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 499, in _run_named_probe\n return _probe_dot("e4m3", "e5m2")\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 425, in _probe_dot\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.25, "dtype": "torch.float32", "finite": true, "max_abs_error": 0.64825439453125, "mean_abs_error": 0.25658777356147766, "rtol": 0.1, "shape": [16, 16]}\n"
},
{
"duration_seconds": 2.881,
"error": "{"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1947364807128906, "mean_abs_error": 0.5505611300468445, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}",
"error_type": "AssertionError",
"name": "dot_scaled_e4m3",
"returncode": 1,
"status": "FAIL",
"traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 501, in _run_named_probe\n return probe_dot_scaled(name.removeprefix("dot_scaled"))\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 458, in _probe_dot_scaled\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1947364807128906, "mean_abs_error": 0.5505611300468445, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}\n"
},
{
"duration_seconds": 2.857,
"error": "{"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1368446350097656, "mean_abs_error": 0.5454071760177612, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}",
"error_type": "AssertionError",
"name": "dot_scaled_e5m2",
"returncode": 1,
"status": "FAIL",
"traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 501, in _run_named_probe\n return probe_dot_scaled(name.removeprefix("dot_scaled"))\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 458, in _probe_dot_scaled\n raise AssertionError(json.dumps(result, sort_keys=True))\nAssertionError: {"allclose": false, "atol": 0.5, "dtype": "torch.float32", "finite": true, "max_abs_error": 1.1368446350097656, "mean_abs_error": 0.5454071760177612, "rtol": 0.15, "scale_contract": "MXFP8 UE8M0; encoding 127 assumed to mean 1.0", "shape": [16, 16]}\n"
},
{
"duration_seconds": 2.688,
"error": "_scaled_mm:../third_party/op-plugin/op_plugin/ops/opapi/ScaledMmKernelNpuOpApi.cpp:433 NPU function error: call aclnnQuantMatmulV5 failed, error code is 161002\n[ERROR] 2026-07-31-08:55:21 (PID:1494979, Device:0, RankID:-1) ERR00100 PTA call acl api failed.\n[PID: 1494979] 2026-07-31-08:55:21.388.591 Invalid_Argument_Tensor_Dtype(EZ0020): Parameter out of aclnnQuantMatmulV5GetWorkspaceSize has incorrect dtype DT_INT8. Reason: the dtype of out must be in dtype support list [DT_FLOAT16,DT_BFLOAT16,DT_FLOAT,].\n Solution: Check whether the dtype of input/output tensor meets the condition.\n",
"error_type": "RuntimeError",
"name": "torch_scaled_mm_e4m3",
"returncode": 1,
"status": "FAIL",
"traceback": "Traceback (most recent call last):\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 511, in _child_main\n details = _run_named_probe(name)\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 503, in _run_named_probe\n return _probe_torch_scaled_mm()\n File "/home/l30073763/fp8/fp8_triton_ascend_prob.py", line 469, in _probe_torch_scaled_mm\n out = torch._scaled_mm(a, b, scale_a, scale_b)\nRuntimeError: _scaled_mm:../third_party/op-plugin/op_plugin/ops/opapi/ScaledMmKernelNpuOpApi.cpp:433 NPU function error: call aclnnQuantMatmulV5 failed, error code is 161002\n[ERROR] 2026-07-31-08:55:21 (PID:1494979, Device:0, RankID:-1) ERR00100 PTA call acl api failed.\n[PID: 1494979] 2026-07-31-08:55:21.388.591 Invalid_Argument_Tensor_Dtype(EZ0020): Parameter out of aclnnQuantMatmulV5GetWorkspaceSize has incorrect dtype DT_INT8. Reason: the dtype of out must be in dtype support list [DT_FLOAT16,DT_BFLOAT16,DT_FLOAT,].\n Solution: Check whether the dtype of input/output tensor meets the condition.\n\n"
}
],
"script": "/home/l30073763/fp8/fp8_triton_ascend_prob.py",
"timeout_per_probe_seconds": 180
}
FP8_PROBE_SUMMARY_END
JSON summary written to: /home/l30073763/fp8/fp8_probe_a5.json