Generate baseline data script for operator with single input #78
Conversation
BruceDai
changed the title
BruceDai
changed the title
| ``` | ||
|
|
||
| then, you can find two generated folders named 'test-data' and | ||
| 'test-data-wpt'. There're raw test data as being |
| node gen-operator-with-single-input.js resources\softsign.json | ||
| ``` | ||
|
|
||
| then, you can find two generated folders named 'test-data' and |
| }; | ||
| const inputTensor = new Tensor(input.shape, input.data); | ||
| const outputTensor = | ||
| operatorMappingDict[operatorName](inputTensor, options); |
There was a problem hiding this comment.
(minor style) Hmm, code is more readable when it's not awkwardly split like this for going just one column past 80.
const inputTensor = new Tensor(input.shape, input.data);
const outputTensor = operatorMappingDict[operatorName](inputTensor, options);
There's plenty of screen space on monitors in the year 2024, and I'm curious why 80 is being applied here when the root eslint is set to 100 columns? https://github.com/webmachinelearning/webnn-baseline/blob/main/.eslintrc.js#L17 In any case, it seems inconsistent.
| "inputs": { | ||
| "input": { | ||
| "shape": [2, 1, 4, 1, 3], | ||
| "data": "float64Data", |
There was a problem hiding this comment.
| "data": "float64Data", | |
| "data": "float16Data", |
Typo?
The output below is float16Data.
"expected": {
"name": "output",
"shape": [2, 1, 4, 1, 3],
"data": "float16Data",
"type": "float16"
}
| "expected": { | ||
| "name": "output", | ||
| "shape": [24], | ||
| "data": "float321DNegativeDefault", |
There was a problem hiding this comment.
floating point 321-dimensional data? 😉 Propose putting a separator between them, like "float32-1D" or "float32_1D" for some clarity.
|
|
||
| /** | ||
| * Prepare input data by specified config of inputsDataInfo, dataFile, min, | ||
| * max parameters . |
There was a problem hiding this comment.
| * max parameters . | |
| * max parameters. |
| } | ||
| const content = fs.readFileSync(inputFile).toString(); | ||
| const jsonDict = JSON.parse( | ||
| content.replace(/\\"|"(?:\\"|[^"])*"|(\/\/.*|\/\*[\s\S]*?\*\/)/g, |
There was a problem hiding this comment.
What is this regex doing? Is it removing comments?
| content.replace(/\\"|"(?:\\"|[^"])*"|(\/\/.*|\/\*[\s\S]*?\*\/)/g, | |
| content.replace(/\\"|"(?:\\"|[^"])*"|(\/\/.*|\/\*[\s\S]*?\*\/)/g, // remove comments |
| // If found, it replaces it by a trio | ||
| if ( value instanceof Int8Array || | ||
| value instanceof Uint8Array || | ||
| value instanceof Uint16Array || |
There was a problem hiding this comment.
<minor>Should we add Int16Array here right now for consistency with Uint16Array? WebNN doesn't support either int16 or uint16 currently, but if it does in the future, that's one less place in the test code to find a bug later. Though I suppose you are probably adding this for the sake of float16.</minor>
| value instanceof Float32Array || | ||
| value instanceof Float16Array || | ||
| value instanceof Float64Array ) { | ||
| if (value.length ===1) { // value instanceof Uint8Array && |
| const result = []; | ||
| result[0] = value[0]; | ||
| return result; | ||
| } else { |
There was a problem hiding this comment.
Is this the BigInt64Array case? If so, comment is nice.
|
Just noticed this is still open. Is it obviated by other changes? |
|
Should we complete or abandon this? |
This PR includes
softsignandgeluoperators.@huningxin @fdwr PTAL, thanks.