mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-11-26 22:32:46 +00:00
Bug 1929596 - Add MLSuggest inference as under ml tests and make it run inference?tarek,adw,sparky r=urlbar-reviewers,perftest-reviewers,sparky,adw,tarek
Differential Revision: https://phabricator.services.mozilla.com/D228192
This commit is contained in:
Chidam Gopal
parent
9f26ffa179
commit
485c1c07d3
@ -13,24 +13,6 @@ ChromeUtils.defineESModuleGetters(lazy, {
|
||||
UrlbarPrefs: "resource:///modules/UrlbarPrefs.sys.mjs",
|
||||
});
|
||||
|
||||
/**
|
||||
* These INTENT_OPTIONS and NER_OPTIONS will go to remote setting server and depends
|
||||
* on https://bugzilla.mozilla.org/show_bug.cgi?id=1923553
|
||||
*/
|
||||
const INTENT_OPTIONS = {
|
||||
taskName: "text-classification",
|
||||
featureId: "suggest-intent-classification",
|
||||
engineId: "ml-suggest-intent",
|
||||
timeoutMS: -1,
|
||||
};
|
||||
|
||||
const NER_OPTIONS = {
|
||||
taskName: "token-classification",
|
||||
featureId: "suggest-NER",
|
||||
engineId: "ml-suggest-ner",
|
||||
timeoutMS: -1,
|
||||
};
|
||||
|
||||
// List of prepositions used in subject cleaning.
|
||||
const PREPOSITIONS = ["in", "at", "on", "for", "to", "near"];
|
||||
|
||||
@ -42,6 +24,20 @@ const PREPOSITIONS = ["in", "at", "on", "for", "to", "near"];
|
||||
class _MLSuggest {
|
||||
#modelEngines = {};
|
||||
|
||||
INTENT_OPTIONS = {
|
||||
taskName: "text-classification",
|
||||
featureId: "suggest-intent-classification",
|
||||
engineId: "ml-suggest-intent",
|
||||
timeoutMS: -1,
|
||||
};
|
||||
|
||||
NER_OPTIONS = {
|
||||
taskName: "token-classification",
|
||||
featureId: "suggest-NER",
|
||||
engineId: "ml-suggest-ner",
|
||||
timeoutMS: -1,
|
||||
};
|
||||
|
||||
// Helper to wrap createEngine for testing purpose
|
||||
createEngine(args) {
|
||||
return lazy.createEngine(args);
|
||||
@ -52,8 +48,8 @@ class _MLSuggest {
|
||||
*/
|
||||
async initialize() {
|
||||
await Promise.all([
|
||||
this.#initializeModelEngine(INTENT_OPTIONS),
|
||||
this.#initializeModelEngine(NER_OPTIONS),
|
||||
this.#initializeModelEngine(this.INTENT_OPTIONS),
|
||||
this.#initializeModelEngine(this.NER_OPTIONS),
|
||||
]);
|
||||
}
|
||||
|
||||
@ -99,10 +95,10 @@ class _MLSuggest {
|
||||
);
|
||||
|
||||
return {
|
||||
intent: intentRes,
|
||||
intent: intentRes[0].label,
|
||||
location: locationResVal,
|
||||
subject: this.#findSubjectFromQuery(query, locationResVal),
|
||||
metrics: this.#sumObjectsByKey(intentRes.metrics, nerResult.metrics),
|
||||
metrics: { intent: intentRes.metrics, ner: nerResult.metrics },
|
||||
};
|
||||
}
|
||||
|
||||
@ -142,11 +138,12 @@ class _MLSuggest {
|
||||
* The user's input query.
|
||||
* @param {object} options
|
||||
* The options for the engine pipeline
|
||||
* @returns {string|null}
|
||||
* The predicted intent label or null if the model is not initialized.
|
||||
* @returns {object[] | null}
|
||||
* The intent results or null if the model is not initialized.
|
||||
*/
|
||||
async _findIntent(query, options = {}) {
|
||||
const engineIntentClassifier = this.#modelEngines[INTENT_OPTIONS.engineId];
|
||||
const engineIntentClassifier =
|
||||
this.#modelEngines[this.INTENT_OPTIONS.engineId];
|
||||
if (!engineIntentClassifier) {
|
||||
return null;
|
||||
}
|
||||
@ -160,12 +157,11 @@ class _MLSuggest {
|
||||
} catch (error) {
|
||||
// engine could timeout or fail, so remove that from cache
|
||||
// and reinitialize
|
||||
this.#modelEngines[INTENT_OPTIONS.engineId] = null;
|
||||
this.#initializeModelEngine(INTENT_OPTIONS);
|
||||
this.#modelEngines[this.INTENT_OPTIONS.engineId] = null;
|
||||
this.#initializeModelEngine(this.INTENT_OPTIONS);
|
||||
return null;
|
||||
}
|
||||
// Return the first label from the result
|
||||
return res[0].label;
|
||||
return res;
|
||||
}
|
||||
|
||||
/**
|
||||
@ -180,14 +176,14 @@ class _MLSuggest {
|
||||
* The NER results or null if the model is not initialized.
|
||||
*/
|
||||
async _findNER(query, options = {}) {
|
||||
const engineNER = this.#modelEngines[NER_OPTIONS.engineId];
|
||||
const engineNER = this.#modelEngines[this.NER_OPTIONS.engineId];
|
||||
try {
|
||||
return engineNER?.run({ args: [query], options });
|
||||
} catch (error) {
|
||||
// engine could timeout or fail, so remove that from cache
|
||||
// and reinitialize
|
||||
this.#modelEngines[NER_OPTIONS.engineId] = null;
|
||||
this.#initializeModelEngine(NER_OPTIONS);
|
||||
this.#modelEngines[this.NER_OPTIONS.engineId] = null;
|
||||
this.#initializeModelEngine(this.NER_OPTIONS);
|
||||
return null;
|
||||
}
|
||||
}
|
||||
@ -307,17 +303,6 @@ class _MLSuggest {
|
||||
}
|
||||
return subject;
|
||||
}
|
||||
|
||||
#sumObjectsByKey(...objs) {
|
||||
return objs.reduce((a, b) => {
|
||||
for (let k in b) {
|
||||
if (b.hasOwnProperty(k)) {
|
||||
a[k] = (a[k] || 0) + b[k];
|
||||
}
|
||||
}
|
||||
return a;
|
||||
}, {});
|
||||
}
|
||||
}
|
||||
|
||||
// Export the singleton instance
|
||||
|
||||
@ -59,6 +59,7 @@ suites:
|
||||
tests:
|
||||
"ML Test Model": ""
|
||||
"ML Test Multi Model": ""
|
||||
"ML Suggest Inference Model": ""
|
||||
"ML Suggest Intent Model": ""
|
||||
"ML Suggest NER Model": ""
|
||||
|
||||
|
||||
@ -444,3 +444,79 @@ ml-multi-perf:
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_engine_multi_perf.js
|
||||
|
||||
ml-perf-suggest-inf:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Inference Model
|
||||
treeherder:
|
||||
symbol: perftest(linux-ml-perf-suggest-inf)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox-bin
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_inference.js
|
||||
|
||||
ml-perf-suggest-int:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Intent Model
|
||||
treeherder:
|
||||
symbol: perftest(linux-ml-perf-suggest-int)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox-bin
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_intent_perf.js
|
||||
|
||||
ml-perf-suggest-ner:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
description: Run ML Suggest NER Model
|
||||
treeherder:
|
||||
symbol: perftest(linux-ml-perf-suggest-ner)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox-bin
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_ner_perf.js
|
||||
|
||||
@ -391,3 +391,79 @@ ml-multi-perf:
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_engine_multi_perf.js
|
||||
|
||||
ml-perf-suggest-inf:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Inference Model
|
||||
treeherder:
|
||||
symbol: perftest(mac-ml-perf-suggest-inf)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/target.dmg
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_inference.js
|
||||
|
||||
ml-perf-suggest-int:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Intent Model
|
||||
treeherder:
|
||||
symbol: perftest(mac-ml-perf-suggest-int)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/target.dmg
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_intent_perf.js
|
||||
|
||||
ml-perf-suggest-ner:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
description: Run ML Suggest NER Model
|
||||
treeherder:
|
||||
symbol: perftest(mac-ml-perf-suggest-ner)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/target.dmg
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_ner_perf.js
|
||||
|
||||
@ -319,3 +319,79 @@ ml-multi-perf:
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_engine_multi_perf.js
|
||||
|
||||
ml-perf-suggest-inf:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Inference Model
|
||||
treeherder:
|
||||
symbol: perftest(win-ml-perf-suggest-inf)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox.exe
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_inference.js
|
||||
|
||||
ml-perf-suggest-int:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-intent
|
||||
description: Run ML Suggest Intent Model
|
||||
treeherder:
|
||||
symbol: perftest(win-ml-perf-suggest-int)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox.exe
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_intent_perf.js
|
||||
|
||||
ml-perf-suggest-ner:
|
||||
fetches:
|
||||
fetch:
|
||||
- ort.wasm
|
||||
- ort.jsep.wasm
|
||||
- ort-training.wasm
|
||||
- mozilla-ner
|
||||
description: Run ML Suggest NER Model
|
||||
treeherder:
|
||||
symbol: perftest(win-ml-perf-suggest-ner)
|
||||
tier: 2
|
||||
attributes:
|
||||
batch: false
|
||||
cron: false
|
||||
run-on-projects: []
|
||||
run:
|
||||
command: >-
|
||||
mkdir -p $MOZ_FETCHES_DIR/../artifacts &&
|
||||
cd $MOZ_FETCHES_DIR &&
|
||||
python3 python/mozperftest/mozperftest/runner.py
|
||||
--mochitest-binary ${MOZ_FETCHES_DIR}/firefox/firefox.exe
|
||||
--flavor mochitest
|
||||
--output $MOZ_FETCHES_DIR/../artifacts
|
||||
toolkit/components/ml/tests/browser/browser_ml_suggest_ner_perf.js
|
||||
|
||||
@ -391,6 +391,24 @@ toolkit/components/ml/tests/browser
|
||||
-----------------------------------
|
||||
Performance tests running through Mochitest for ML Models
|
||||
|
||||
browser_ml_suggest_inference.js
|
||||
===============================
|
||||
|
||||
:owner: GenAI Team
|
||||
:name: ML Suggest Inference Model
|
||||
:Default options:
|
||||
|
||||
::
|
||||
|
||||
--perfherder
|
||||
--perfherder-metrics name:inference-pipeline-ready-latency,unit:ms,shouldAlert:True, name:inference-initialization-latency,unit:ms,shouldAlert:True, name:inference-model-run-latency,unit:ms,shouldAlert:True, name:inference-pipeline-ready-memory,unit:MB,shouldAlert:True, name:inference-initialization-memory,unit:MB,shouldAlert:True, name:inference-model-run-memory,unit:MB,shouldAlert:True
|
||||
--verbose
|
||||
--manifest perftest.toml
|
||||
--manifest-flavor browser-chrome
|
||||
--try-platform linux, mac, win
|
||||
|
||||
**Template test for ML suggest Inference Model**
|
||||
|
||||
browser_ml_suggest_intent_perf.js
|
||||
=================================
|
||||
|
||||
|
||||
@ -0,0 +1,205 @@
|
||||
/* Any copyright is dedicated to the Public Domain.
|
||||
http://creativecommons.org/publicdomain/zero/1.0/ */
|
||||
"use strict";
|
||||
|
||||
const ITERATIONS = 1;
|
||||
|
||||
const PREFIX = "inference";
|
||||
const METRICS = [
|
||||
`${PREFIX}-${PIPELINE_READY_LATENCY}`,
|
||||
`${PREFIX}-${INITIALIZATION_LATENCY}`,
|
||||
`${PREFIX}-${MODEL_RUN_LATENCY}`,
|
||||
`${PREFIX}-${PIPELINE_READY_MEMORY}`,
|
||||
`${PREFIX}-${INITIALIZATION_MEMORY}`,
|
||||
`${PREFIX}-${MODEL_RUN_MEMORY}`,
|
||||
];
|
||||
const journal = {};
|
||||
for (let metric of METRICS) {
|
||||
journal[metric] = [1];
|
||||
}
|
||||
|
||||
const perfMetadata = {
|
||||
owner: "GenAI Team",
|
||||
name: "ML Suggest Inference Model",
|
||||
description: "Template test for ML suggest Inference Model",
|
||||
options: {
|
||||
default: {
|
||||
perfherder: true,
|
||||
perfherder_metrics: [
|
||||
{
|
||||
name: "inference-pipeline-ready-latency",
|
||||
unit: "ms",
|
||||
shouldAlert: true,
|
||||
},
|
||||
{
|
||||
name: "inference-initialization-latency",
|
||||
unit: "ms",
|
||||
shouldAlert: true,
|
||||
},
|
||||
{ name: "inference-model-run-latency", unit: "ms", shouldAlert: true },
|
||||
{
|
||||
name: "inference-pipeline-ready-memory",
|
||||
unit: "MB",
|
||||
shouldAlert: true,
|
||||
},
|
||||
{
|
||||
name: "inference-initialization-memory",
|
||||
unit: "MB",
|
||||
shouldAlert: true,
|
||||
},
|
||||
{ name: "inference-model-run-memory", unit: "MB", shouldAlert: true },
|
||||
],
|
||||
verbose: true,
|
||||
manifest: "perftest.toml",
|
||||
manifest_flavor: "browser-chrome",
|
||||
try_platform: ["linux", "mac", "win"],
|
||||
},
|
||||
},
|
||||
};
|
||||
|
||||
requestLongerTimeout(120);
|
||||
|
||||
const CUSTOM_INTENT_OPTIONS = {
|
||||
taskName: "text-classification",
|
||||
featureId: "suggest-intent-classification",
|
||||
engineId: "ml-suggest-intent",
|
||||
timeoutMS: -1,
|
||||
modelId: "Mozilla/mobilebert-uncased-finetuned-LoRA-intent-classifier",
|
||||
dtype: "q8",
|
||||
modelRevision: "main",
|
||||
};
|
||||
|
||||
const CUSTOM_NER_OPTIONS = {
|
||||
taskName: "token-classification",
|
||||
featureId: "suggest-NER",
|
||||
engineId: "ml-suggest-ner",
|
||||
timeoutMS: -1,
|
||||
modelId: "Mozilla/distilbert-uncased-NER-LoRA",
|
||||
dtype: "q8",
|
||||
modelRevision: "main",
|
||||
};
|
||||
|
||||
const ROOT_URL =
|
||||
"chrome://mochitests/content/browser/toolkit/components/ml/tests/browser/data/suggest";
|
||||
const YELP_KEYWORDS_DATA = `${ROOT_URL}/yelp_val_keywords_data.json`;
|
||||
const YELP_VAL_DATA = `${ROOT_URL}/yelp_val_generated_data.json`;
|
||||
const NER_VAL_DATA = `${ROOT_URL}/named_entity_val_generated_data.json`;
|
||||
|
||||
async function get_val_data(inputDataPath) {
|
||||
const response = await fetch(inputDataPath);
|
||||
if (!response.ok) {
|
||||
throw new Error(
|
||||
`Failed to fetch data: ${response.statusText} from ${inputDataPath}`
|
||||
);
|
||||
}
|
||||
return response.json();
|
||||
}
|
||||
|
||||
async function read_data_by_type(type) {
|
||||
let data;
|
||||
if (type === "YELP_KEYWORDS_DATA") {
|
||||
data = await get_val_data(YELP_KEYWORDS_DATA);
|
||||
return data[0].subjects;
|
||||
} else if (type === "YELP_VAL_DATA") {
|
||||
data = await get_val_data(YELP_VAL_DATA);
|
||||
return data.queries;
|
||||
} else if (type === "NER_VAL_DATA") {
|
||||
data = await get_val_data(NER_VAL_DATA);
|
||||
return data.queries;
|
||||
}
|
||||
return [];
|
||||
}
|
||||
|
||||
// Utility to write results to a local JSON file using IOUtils
|
||||
async function writeResultsToFile(results, type) {
|
||||
try {
|
||||
const json = JSON.stringify(results, null, 2);
|
||||
const OUTPUT_FILE_PATH = `${
|
||||
Services.dirsvc.get("DfltDwnld", Ci.nsIFile).path
|
||||
}/ML_output_${type}.json`;
|
||||
await IOUtils.writeUTF8(OUTPUT_FILE_PATH, json);
|
||||
console.log("Results successfully written to:", OUTPUT_FILE_PATH);
|
||||
} catch (error) {
|
||||
console.error("Failed to write results to file:", error);
|
||||
Assert.ok(false, "Failed to write results to file");
|
||||
}
|
||||
}
|
||||
|
||||
async function perform_inference(queries, type) {
|
||||
// Ensure MLSuggest is initialized
|
||||
await MLSuggest.initialize();
|
||||
|
||||
const batchSize = 32;
|
||||
const results = [];
|
||||
|
||||
// Process in batches of 32
|
||||
for (let i = 0; i < queries.length; i += batchSize) {
|
||||
const batchQueries = queries.slice(i, i + batchSize);
|
||||
const batchResults = await Promise.all(
|
||||
batchQueries.map(async query => {
|
||||
const suggestion = await MLSuggest.makeSuggestions(query);
|
||||
const res = {
|
||||
query,
|
||||
intent: suggestion.intent,
|
||||
city: suggestion.location.city,
|
||||
state: suggestion.location.state,
|
||||
};
|
||||
return res;
|
||||
})
|
||||
);
|
||||
results.push(...batchResults);
|
||||
}
|
||||
Assert.ok(
|
||||
results.length === queries.length,
|
||||
"results size should be equal to queries size."
|
||||
);
|
||||
// Write results to a file
|
||||
await writeResultsToFile(results, type);
|
||||
}
|
||||
|
||||
const runInference2 = async () => {
|
||||
ChromeUtils.defineESModuleGetters(this, {
|
||||
MLSuggest: "resource:///modules/urlbar/private/MLSuggest.sys.mjs",
|
||||
});
|
||||
|
||||
// Override INTENT and NER options within MLSuggest
|
||||
MLSuggest.INTENT_OPTIONS = CUSTOM_INTENT_OPTIONS;
|
||||
MLSuggest.NER_OPTIONS = CUSTOM_NER_OPTIONS;
|
||||
|
||||
const modelDirectory = normalizePathForOS(
|
||||
`${Services.env.get("MOZ_FETCHES_DIR")}/onnx-models`
|
||||
);
|
||||
info(`Model Directory: ${modelDirectory}`);
|
||||
const { baseUrl: modelHubRootUrl } = startHttpServer(modelDirectory);
|
||||
info(`ModelHubRootUrl: ${modelHubRootUrl}`);
|
||||
const { cleanup } = await perfSetup({
|
||||
prefs: [
|
||||
["browser.ml.modelHubRootUrl", modelHubRootUrl],
|
||||
["javascript.options.wasm_lazy_tiering", true],
|
||||
],
|
||||
});
|
||||
|
||||
const TYPES = ["YELP_KEYWORDS_DATA", "YELP_VAL_DATA", "NER_VAL_DATA"];
|
||||
for (const type of TYPES) {
|
||||
info(`processing ${type} now`);
|
||||
// Read data for the current type
|
||||
const queries = await read_data_by_type(type);
|
||||
if (!queries) {
|
||||
info(`No queries found for type: ${type}`);
|
||||
continue;
|
||||
}
|
||||
// Run inference for each query
|
||||
await perform_inference(queries, type);
|
||||
}
|
||||
await MLSuggest.shutdown();
|
||||
await EngineProcess.destroyMLEngine();
|
||||
await cleanup();
|
||||
};
|
||||
|
||||
/**
|
||||
* Tests remote ml model
|
||||
*/
|
||||
add_task(async function test_ml_generic_pipeline() {
|
||||
await runInference2();
|
||||
reportMetrics(journal);
|
||||
});
|
||||
@ -6,12 +6,12 @@ const ITERATIONS = 10;
|
||||
|
||||
const PREFIX = "intent";
|
||||
const METRICS = [
|
||||
`${PREFIX}_${PIPELINE_READY_LATENCY}`,
|
||||
`${PREFIX}_${INITIALIZATION_LATENCY}`,
|
||||
`${PREFIX}_${MODEL_RUN_LATENCY}`,
|
||||
`${PREFIX}_${PIPELINE_READY_MEMORY}`,
|
||||
`${PREFIX}_${INITIALIZATION_MEMORY}`,
|
||||
`${PREFIX}_${MODEL_RUN_MEMORY}`,
|
||||
`${PREFIX}-${PIPELINE_READY_LATENCY}`,
|
||||
`${PREFIX}-${INITIALIZATION_LATENCY}`,
|
||||
`${PREFIX}-${MODEL_RUN_LATENCY}`,
|
||||
`${PREFIX}-${PIPELINE_READY_MEMORY}`,
|
||||
`${PREFIX}-${INITIALIZATION_MEMORY}`,
|
||||
`${PREFIX}-${MODEL_RUN_MEMORY}`,
|
||||
];
|
||||
const journal = {};
|
||||
for (let metric of METRICS) {
|
||||
@ -55,6 +55,19 @@ requestLongerTimeout(120);
|
||||
* Tests local suggest intent model
|
||||
*/
|
||||
add_task(async function test_ml_generic_pipeline() {
|
||||
const modelDirectory = normalizePathForOS(
|
||||
`${Services.env.get("MOZ_FETCHES_DIR")}/onnx-models`
|
||||
);
|
||||
info(`Model Directory: ${modelDirectory}`);
|
||||
const { baseUrl: modelHubRootUrl } = startHttpServer(modelDirectory);
|
||||
info(`ModelHubRootUrl: ${modelHubRootUrl}`);
|
||||
const { cleanup } = await perfSetup({
|
||||
prefs: [
|
||||
["browser.ml.modelHubRootUrl", modelHubRootUrl],
|
||||
["javascript.options.wasm_lazy_tiering", true],
|
||||
],
|
||||
});
|
||||
|
||||
const options = new PipelineOptions({
|
||||
taskName: "text-classification",
|
||||
modelId: "Mozilla/mobilebert-uncased-finetuned-LoRA-intent-classifier",
|
||||
@ -68,9 +81,12 @@ add_task(async function test_ml_generic_pipeline() {
|
||||
for (let i = 0; i < ITERATIONS; i++) {
|
||||
let metrics = await runInference(options, args);
|
||||
for (let [metricName, metricVal] of Object.entries(metrics)) {
|
||||
if (metricName !== `${MODEL_RUN_MEMORY}`) {
|
||||
Assert.ok(metricVal >= 0, "Metric should be non-negative.");
|
||||
journal[`${PREFIX}_${metricName}`].push(metricVal);
|
||||
}
|
||||
journal[`${PREFIX}-${metricName}`].push(metricVal);
|
||||
}
|
||||
}
|
||||
reportMetrics(journal);
|
||||
await cleanup();
|
||||
});
|
||||
|
||||
@ -6,12 +6,12 @@ const ITERATIONS = 10;
|
||||
|
||||
const PREFIX = "NER";
|
||||
const METRICS = [
|
||||
`${PREFIX}_${PIPELINE_READY_LATENCY}`,
|
||||
`${PREFIX}_${INITIALIZATION_LATENCY}`,
|
||||
`${PREFIX}_${MODEL_RUN_LATENCY}`,
|
||||
`${PREFIX}_${PIPELINE_READY_MEMORY}`,
|
||||
`${PREFIX}_${INITIALIZATION_MEMORY}`,
|
||||
`${PREFIX}_${MODEL_RUN_MEMORY}`,
|
||||
`${PREFIX}-${PIPELINE_READY_LATENCY}`,
|
||||
`${PREFIX}-${INITIALIZATION_LATENCY}`,
|
||||
`${PREFIX}-${MODEL_RUN_LATENCY}`,
|
||||
`${PREFIX}-${PIPELINE_READY_MEMORY}`,
|
||||
`${PREFIX}-${INITIALIZATION_MEMORY}`,
|
||||
`${PREFIX}-${MODEL_RUN_MEMORY}`,
|
||||
];
|
||||
const journal = {};
|
||||
for (let metric of METRICS) {
|
||||
@ -47,6 +47,19 @@ requestLongerTimeout(120);
|
||||
* Tests local suggest NER model
|
||||
*/
|
||||
add_task(async function test_ml_generic_pipeline() {
|
||||
const modelDirectory = normalizePathForOS(
|
||||
`${Services.env.get("MOZ_FETCHES_DIR")}/onnx-models`
|
||||
);
|
||||
info(`Model Directory: ${modelDirectory}`);
|
||||
const { baseUrl: modelHubRootUrl } = startHttpServer(modelDirectory);
|
||||
info(`ModelHubRootUrl: ${modelHubRootUrl}`);
|
||||
const { cleanup } = await perfSetup({
|
||||
prefs: [
|
||||
["browser.ml.modelHubRootUrl", modelHubRootUrl],
|
||||
["javascript.options.wasm_lazy_tiering", true],
|
||||
],
|
||||
});
|
||||
|
||||
const options = new PipelineOptions({
|
||||
taskName: "token-classification",
|
||||
modelId: "Mozilla/distilbert-uncased-NER-LoRA",
|
||||
@ -61,8 +74,9 @@ add_task(async function test_ml_generic_pipeline() {
|
||||
let metrics = await runInference(options, args);
|
||||
for (let [metricName, metricVal] of Object.entries(metrics)) {
|
||||
Assert.ok(metricVal >= 0, "Metric should be non-negative.");
|
||||
journal[`${PREFIX}_${metricName}`].push(metricVal);
|
||||
journal[`${PREFIX}-${metricName}`].push(metricVal);
|
||||
}
|
||||
}
|
||||
reportMetrics(journal);
|
||||
await cleanup();
|
||||
});
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@ -337,6 +337,7 @@ async function perfSetup({ disabled = false, prefs = [] } = {}) {
|
||||
100,
|
||||
200
|
||||
);
|
||||
await SpecialPowers.popPrefEnv();
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
@ -9,6 +9,9 @@ disabled = "Disabled as we want to run this only as perftest, not regular CI"
|
||||
["browser_ml_engine_perf.js"]
|
||||
disabled = "Disabled as we want to run this only as perftest, not regular CI"
|
||||
|
||||
["browser_ml_suggest_inference.js"]
|
||||
disabled = "Disabled as we want to run this only as perftest, not regular CI"
|
||||
|
||||
["browser_ml_suggest_intent_perf.js"]
|
||||
disabled = "Disabled as we want to run this only as perftest, not regular CI"
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user