FAQ

[toc]

Self-Check Before Debugging

Is TTK up to date? Run python3 -m ttk -v
Check command syntax: python3 -m ttk kernel --help
Is CANN installed and environment sourced?
Check operator name spelling in CSV
If using custom plugins, verify registration names match CSV operator names

Environment Issues

"ASCEND_HOME_PATH not found"

Install CANN and source environment:

source /usr/local/Ascend/ascend-toolkit/set_env.sh

CANN Sub-Package Environment

source /usr/local/Ascend/latest/bin/setenv.bash
export ASCEND_HOME_PATH=/usr/local/Ascend/latest
export ASCEND_TOOLKIT_HOME=/usr/local/Ascend/latest

Common errors without sourcing: ImportError: libhccl.so, AttributeError: 'NoneType' object has no attribute 'acl_init'

CSV Writing Issues

How to write dtype strings?

Supported: float16/fp16, float32/fp32, float64/fp64, bfloat16/bf16, int8, int16, int32, int64, uint8, uint16, uint32, uint64, bool, complex64, complex128.

Fields with special characters must be double-quoted

"((128, 1024), (1, 1024))"     # Correct
((128, 1024), (1, 1024))       # Wrong - commas parsed as CSV delimiters

Output shape depends on input tensor values, cannot be inferred from input shapes (e.g. NonZero)

Set output_shapes to a placeholder shape (non-empty), and mark the unknown output index in output_shape_unknown_indexes:

output_shapes,"((8, 1716),)"
output_shape_unknown_indexes,"(0,)"

The framework will determine the actual output shape at runtime.

TensorList operators

Express TensorList via nesting in input_shapes. Group tensors in the same inner tuple:

input_shapes,"(((3,3),(3,2),(3,4)),(3,5))"

Preview CSV test cases

python3 -m ttk list -i cases.csv
python3 -m ttk list -i cases.csv --op add

Execution Issues

Timeout or OOM with large operators

--proc-timeout=300 (per-case timeout)
-l 10 (10GB HBM limit)
--reserve-hbm=512 (reserve HBM)

Reproduce test results

python3 -m ttk kernel -i cases.csv --seed 42

Quick environment check

python3 -m ttk info
python3 -m ttk kernel -i examples/case_store/kernel/add.csv

Precision Issues

How to debug precision failures?

python3 -m ttk kernel -i cases.csv --dump-on-fail
python3 -m ttk kernel -i cases.csv --dump in,golden --dump-format npy
python3 -m ttk kernel -i cases.csv -t add_01 --single-log --dump full

Comparison method recommendations

Scenario	Method	Flag
General float (default)	Statistical relative error (community standard)	`--compare stat_rel_err` (default)
Pointwise isclose	Numeric approximation	`--compare close`
Large vector trends	Cosine similarity	`--compare cosine`
Integer / exact match	Binary exact	`--compare binary`
float8 types	Requantization	`--compare requant` (auto)
3rd-party cross-check	Three-party cross-check	`--compare cross_check` (needs `third_party`)

Adjust tolerance in CSV

precision_tolerances,"((0.001, 0.001),)"
absolute_precision,1e-8

Plugin Issues

Custom golden not taking effect

Verify --plugin path is correct
Check the TestSpec class name / __spec__ registry key matches CSV op_name (class name = PascalCase+TestSpec, e.g. AbsTestSpec)
Check plugin file for syntax errors

Duplicate operator name registration

Later-loaded plugins override earlier ones (with a warning). Ensure unique registration names.