# tensorplay.cuda.tunable API

Source: https://www.tensorplay.cn/docs/api/tensorplay.cuda.tunable.html

## Functions 20

[#](#api-tensorplay.cuda.tunable.disable)

### disable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.disable.html)

```python
tensorplay.cuda.tunable.disable() → None
```

Turn GEMM kernel tuning off. See [enable()](/docs/generated/tensorplay.cuda.tunable.enable.html#tensorplay.cuda.tunable.enable).

[#](#api-tensorplay.cuda.tunable.enable)

### enable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.enable.html)

```python
tensorplay.cuda.tunable.enable(val: bool = True) → None
```

Turn GEMM kernel tuning on or off.

While on, every GEMM on the tunable path reuses the algorithm recorded
for its signature (or measures one, when tuning is also enabled and no
winner is recorded yet). While off, dispatch is exactly the untuned
behavior.

[#](#api-tensorplay.cuda.tunable.get_filename)

### get_filename

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.get_filename.html)

```python
tensorplay.cuda.tunable.get_filename() → str
```

Return the configured results filename (empty until set or first use).

[#](#api-tensorplay.cuda.tunable.get_max_tuning_duration)

### get_max_tuning_duration

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.get_max_tuning_duration.html)

```python
tensorplay.cuda.tunable.get_max_tuning_duration() → int
```

Return the per-candidate measurement time limit in milliseconds.

[#](#api-tensorplay.cuda.tunable.get_max_tuning_samples)

### get_max_tuning_samples

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.get_max_tuning_samples.html)

```python
tensorplay.cuda.tunable.get_max_tuning_samples() → int
```

Return the per-candidate sample limit.

[#](#api-tensorplay.cuda.tunable.get_results)

### get_results

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.get_results.html)

```python
tensorplay.cuda.tunable.get_results() → list[tuple[str, str, str, float]]
```

Return every recorded winner as (op, params, kernel, time_ms).

[#](#api-tensorplay.cuda.tunable.is_enabled)

### is_enabled

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.is_enabled.html)

```python
tensorplay.cuda.tunable.is_enabled() → bool
```

Return whether GEMM kernel tuning is enabled.

[#](#api-tensorplay.cuda.tunable.is_verbose)

### is_verbose

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.is_verbose.html)

```python
tensorplay.cuda.tunable.is_verbose() → bool
```

Return whether diagnostic logging is on.

[#](#api-tensorplay.cuda.tunable.read_file)

### read_file

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.read_file.html)

```python
tensorplay.cuda.tunable.read_file(filename: str | None = None) → bool
```

Merge a tuning results file into the in-memory database.

The file’s validator lines must match the current build and device;
otherwise the file is rejected and False is returned. Entries
already measured in this process win over the file’s. If
filename is not given, the configured results file is used.

[#](#api-tensorplay.cuda.tunable.record_untuned_disable)

### record_untuned_disable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.record_untuned_disable.html)

```python
tensorplay.cuda.tunable.record_untuned_disable() → None
```

Stop logging untuned GEMMs. See [record_untuned_enable()](/docs/generated/tensorplay.cuda.tunable.record_untuned_enable.html#tensorplay.cuda.tunable.record_untuned_enable).

[#](#api-tensorplay.cuda.tunable.record_untuned_enable)

### record_untuned_enable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.record_untuned_enable.html)

```python
tensorplay.cuda.tunable.record_untuned_enable(val: bool = True) → None
```

Control logging of GEMMs that ran without a tuned choice.

When enabled, every unique signature that runs untuned is appended to
the untuned file (tunableop_untuned<device>.csv in the working
directory), one line per signature.

[#](#api-tensorplay.cuda.tunable.record_untuned_is_enabled)

### record_untuned_is_enabled

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.record_untuned_is_enabled.html)

```python
tensorplay.cuda.tunable.record_untuned_is_enabled() &#x2192; bool
```

Return whether untuned GEMMs are being logged.

[#](#api-tensorplay.cuda.tunable.set_filename)

### set_filename

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.set_filename.html)

```python
tensorplay.cuda.tunable.set_filename(filename: str, insert_device_ordinal: bool = False) &#x2192; None
```

Set the file used to persist tuning results.

If insert_device_ordinal is True, the current device ordinal
is embedded in the name (a %d token is replaced in place, otherwise
the ordinal lands before the extension). This keeps one-process-per-device
runs from sharing a file. An empty filename turns file persistence off.

[#](#api-tensorplay.cuda.tunable.set_max_tuning_duration)

### set_max_tuning_duration

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.set_max_tuning_duration.html)

```python
tensorplay.cuda.tunable.set_max_tuning_duration(duration_ms: int) &#x2192; None
```

Bound the time spent measuring one candidate, in milliseconds.

Zero disables the limit. When both this and [set_max_tuning_samples()](/docs/generated/tensorplay.cuda.tunable.set_max_tuning_samples.html#tensorplay.cuda.tunable.set_max_tuning_samples)
are set, the smaller bound wins; a measurement always runs at least one
timed sample.

[#](#api-tensorplay.cuda.tunable.set_max_tuning_samples)

### set_max_tuning_samples

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.set_max_tuning_samples.html)

```python
tensorplay.cuda.tunable.set_max_tuning_samples(samples: int) &#x2192; None
```

Bound the timed samples spent measuring one candidate.

Zero disables the limit. When both this and [set_max_tuning_duration()](/docs/generated/tensorplay.cuda.tunable.set_max_tuning_duration.html#tensorplay.cuda.tunable.set_max_tuning_duration)
are set, the smaller bound wins; a measurement always runs at least one
timed sample.

[#](#api-tensorplay.cuda.tunable.set_verbose)

### set_verbose

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.set_verbose.html)

```python
tensorplay.cuda.tunable.set_verbose(val: bool) &#x2192; None
```

Turn diagnostic logging of the tuning context on or off.

Verbose output goes to stderr and reports state changes, measurement
passes and file activity. It is meant for debugging.

[#](#api-tensorplay.cuda.tunable.tuning_disable)

### tuning_disable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.tuning_disable.html)

```python
tensorplay.cuda.tunable.tuning_disable() &#x2192; None
```

Stop measuring untuned GEMM shapes. See [tuning_enable()](/docs/generated/tensorplay.cuda.tunable.tuning_enable.html#tensorplay.cuda.tunable.tuning_enable).

[#](#api-tensorplay.cuda.tunable.tuning_enable)

### tuning_enable

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.tuning_enable.html)

```python
tensorplay.cuda.tunable.tuning_enable(val: bool = True) &#x2192; None
```

Control whether untuned GEMM shapes are measured.

When enabled, a GEMM without a recorded winner times every candidate
algorithm and records the fastest; newly found winners are appended to
the results file as they are measured. When disabled, such a GEMM runs
the library heuristic’s top choice.

[#](#api-tensorplay.cuda.tunable.tuning_is_enabled)

### tuning_is_enabled

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.tuning_is_enabled.html)

```python
tensorplay.cuda.tunable.tuning_is_enabled() &#x2192; bool
```

Return whether untuned GEMM shapes are measured.

[#](#api-tensorplay.cuda.tunable.write_file)

### write_file

function[Full reference ↗](/docs/generated/tensorplay.cuda.tunable.write_file.html)

```python
tensorplay.cuda.tunable.write_file() &#x2192; None
```

Rewrite the results file with every recorded winner.

The file is recreated with fresh validator lines followed by all
in-memory results (both those read from files and those measured in
this process).