[83105] Hugging Face models are marked as Generic instead of Task Classification (#68)

Author: as-suvorov

.github/workflows/check-pr-name.yml

@@ -14,20 +14,26 @@ jobs:
         uses: actions/github-script@v5
         with:
           script: |
+            const prName = context.payload.pull_request.title;
+            const prCreator = context.payload.pull_request.user.login;
             const prNameRegExp = /^(?:\[\d+\]\s)+\w+.*/;
             const skipLabel = '[skip-name]';
-            const prName = context.payload.pull_request.title;
+
             console.log(`Pull Request Name is ${prName}`);
 
-            if (prName.includes(skipLabel)) {
-              console.log('Skipping PR name checks');
+            // Skip the PRs with dependency updates
+            const dependencyBotsNames = ['dependabot', 'snyk-bot'];
+
+            const shouldSkip = prName.includes(skipLabel) || dependencyBotsNames.includes(prCreator);
+            if (shouldSkip) {
+              console.log('Skipping PR name checks.');
               return;
             }
 
             if (!prNameRegExp.test(prName)) {
               console.log('Template: [issue_number] ([another_issue_number] ...) Short description');
-              core.setFailed('Your Pull Request title does not confirm to the template');
+              core.setFailed('Your Pull Request title does not confirm to the template.');
               return;
             }
 
-            console.log('Your Pull Request name confirm to the provided template');
+            console.log('Your Pull Request name confirm to the provided template.');

client/src/app/modules/model-manager/components/card/card.component.ts

@@ -1,4 +1,12 @@
-import { ChangeDetectionStrategy, Component, HostBinding, Input } from '@angular/core';
+import {
+  ChangeDetectionStrategy,
+  Component,
+  EventEmitter,
+  HostBinding,
+  HostListener,
+  Input,
+  Output,
+} from '@angular/core';
 
 @Component({
   selector: 'wb-card',
@@ -8,6 +16,14 @@ import { ChangeDetectionStrategy, Component, HostBinding, Input } from '@angular
 })
 export class CardComponent {
   @HostBinding('class.disabled') @Input() disabled = false;
+
+  @Output() selected = new EventEmitter<void>();
+
+  @HostListener('click') onClick(): void {
+    if (!this.disabled) {
+      this.selected.emit();
+    }
+  }
 }
 
 @Component({

client/src/app/modules/model-manager/components/hugging-face-import-ribbon-content/hugging-face-import-ribbon-content.component.html

@@ -44,7 +44,7 @@
         [disabled]="model.validation.disabled"
         [matTooltip]="model.validation.message"
         [matTooltipDisabled]="!model.validation.disabled"
-        (click)="selectedModel = model"
+        (selected)="selectedModel = model"
         data-test-id="model-card"
       >
         <wb-card-title-row>

client/src/app/modules/model-manager/components/hugging-face-import-ribbon-content/huggingface-model-details/huggingface-model-details.component.html

@@ -6,6 +6,7 @@
 
   <wb-model-zoo-details-parameters>
     <wb-parameter-details class="parameter" *ngFor="let param of parameters" [parameter]="param"></wb-parameter-details>
+    <wb-markdown-text class="hf-card-url" [text]="huggingfaceCardUrl"></wb-markdown-text>
   </wb-model-zoo-details-parameters>
 
   <wb-model-zoo-details-description>

client/src/app/modules/model-manager/components/hugging-face-import-ribbon-content/huggingface-model-details/huggingface-model-details.component.scss

@@ -17,6 +17,11 @@
   }
 }
 
+.hf-card-url {
+  @include wb-text-8();
+  margin-top: 30px;
+}
+
 .readme-not-found {
   color: $brownish-grey;
   margin: auto;

client/src/app/modules/model-manager/components/hugging-face-import-ribbon-content/huggingface-model-details/huggingface-model-details.component.ts

@@ -12,6 +12,8 @@ import { DatePipe } from '@angular/common';
 
 import { Store } from '@ngrx/store';
 
+import { MessagesService } from '@core/services/common/messages.service';
+
 import { ModelDomain, modelDomainNames } from '@store/model-store/model.model';
 import { RootStoreState } from '@store';
 import { HuggingfaceModelStoreActions, HuggingfaceModelStoreSelectors } from '@store/huggingface-model-store';
@@ -35,6 +37,9 @@ export class HuggingfaceModelDetailsComponent {
     this._model = value;
     if (this._model) {
       this.parameters = this._extractHfModelParameters(this._model);
+      this.huggingfaceCardUrl = this._messages.getHint('importHuggingFaceTips', 'huggingfaceModelCard', {
+        id: this._model.id,
+      });
       this._store$.dispatch(HuggingfaceModelStoreActions.loadModelReadme({ huggingfaceModelId: this._model.id }));
     } else {
       this.parameters = null;
@@ -56,12 +61,14 @@ export class HuggingfaceModelDetailsComponent {
   readonly error$ = this._store$.select(HuggingfaceModelStoreSelectors.selectModelReadmeError);
 
   parameters: IParameter[] = [];
+  huggingfaceCardUrl: string;
 
   isImportStarted = false;
 
   constructor(
     private readonly _cdr: ChangeDetectorRef,
     private readonly _store$: Store<RootStoreState.State>,
+    private readonly _messages: MessagesService,
     @Inject(LOCALE_ID) private readonly _localeId: string
   ) {}
 

client/src/app/shared/models/huggingface/huggingface-model.ts

@@ -14,7 +14,6 @@ export interface IHuggingfaceModel {
   lastModified: string;
   tags: string[];
   validation: IHuggingfaceModelValidationResult;
-  siblings?: string[];
   config?: IHuggingfaceModelConfig;
   downloads?: number;
 }

client/src/assets/data/hint-messages.json

@@ -106,7 +106,8 @@
   },
   "importHuggingFaceTips": {
     "externalResourceNotification": "Hugging Face is an external source, there may be connection problems. Contains unverified models, we do not guarantee their performance.",
-    "shownSubsetNotification": "We show only a subset of the models from Hugging Face (Text Classification and PyTorch)."
+    "shownSubsetNotification": "We show only a subset of the models from Hugging Face (Text Classification and PyTorch).",
+    "huggingfaceModelCard": "[Model Card on Hugging Face](https://huggingface.co/${id})"
   },
   "login": {
     "loginTip": "To enter the DL Workbench, use a token that is generated once you start the application. Copy it from the console.\n [Read more](https://docs.openvino.ai/latest/workbench_docs_Workbench_DG_Troubleshooting.html#omz)"

wb/main/huggingface_api/huggingface_api.py

@@ -86,7 +86,6 @@ def json(self):
             'lastModified': self.last_modified,
             'tags': self.tags,
             'validation': self.validation.json(),
-            'siblings': self.siblings,
             'config': self.config.json() if self.config else None,
             'downloads': self.downloads,
         }

wb/main/jobs/create_setup_bundle/create_setup_bundle_job.py

@@ -19,6 +19,8 @@
 import tempfile
 from contextlib import closing
 
+from werkzeug.utils import secure_filename
+
 from wb.extensions_factories.database import get_db_session_for_celery
 from wb.main.enumerates import JobTypesEnum, StatusEnum
 from wb.main.jobs.interfaces.ijob import IJob
@@ -63,14 +65,15 @@ def run(self):
             topology_temporary_path = None
 
             if self.include_model:
-                topology_temporary_path = os.path.join(tmp_scripts_folder, self.topology_name)
+                topology_name = secure_filename(self.topology_name)
+                topology_temporary_path = os.path.join(tmp_scripts_folder, topology_name)
                 os.makedirs(topology_temporary_path)
                 xml_file = find_by_ext(self.topology_path, 'xml')
-                tmp_xml_file = os.path.join(topology_temporary_path, f'{self.topology_name}.xml')
+                tmp_xml_file = os.path.join(topology_temporary_path, f'{topology_name}.xml')
                 shutil.copy(xml_file, tmp_xml_file)
 
                 bin_file = find_by_ext(self.topology_path, 'bin')
-                tmp_bin_file = os.path.join(topology_temporary_path, f'{self.topology_name}.bin')
+                tmp_bin_file = os.path.join(topology_temporary_path, f'{topology_name}.bin')
                 shutil.copy(bin_file, tmp_bin_file)
 
             setup_bundle_creator = SetupBundleCreator(

wb/main/jobs/interfaces/job_observers.py

@@ -48,7 +48,7 @@
                             FileMetaData)
 from wb.main.models.wait_model_upload_job_model import WaitModelUploadJobModel
 from wb.main.utils.observer_pattern import Observer
-from wb.main.utils.utils import get_size_of_files
+from wb.main.utils.utils import get_size_of_files, FileSizeConverter
 
 
 def check_existing_job_model_decorator(func: Callable[['Observer', JobState], None]):
@@ -565,7 +565,7 @@ def update(self, subject_state: JobState):
             file_record.status = StatusEnum.ready
             file_record.write_record(session)
 
-            topology.size = file_record.size
+            topology.size = FileSizeConverter.bytes_to_mb(file_record.size)
             topology.status = StatusEnum.ready
             topology.progress = 100
             topology.write_record(session)

wb/main/pipeline_creators/model_creation/import_huggingface_model_pipeline_creator.py

@@ -16,6 +16,7 @@
 """
 import os
 
+import json
 from sqlalchemy.orm import Session
 
 from config.constants import UPLOAD_FOLDER_MODELS, ORIGINAL_FOLDER
@@ -28,9 +29,11 @@
 from wb.main.models.huggingface.import_huggingface_model_job_model import ImportHuggingfaceJobModel, \
     ImportHuggingfaceJobData
 from wb.main.models.model_optimizer_job_model import ModelOptimizerJobData
+from wb.main.models.topologies_metadata_model import DEFAULT_ACCURACY_CONFIGURATION
 from wb.main.models.topologies_model import ModelJobData
 from wb.main.pipeline_creators.model_creation.base_model_creation_pipeline_creator import \
     BaseModelCreationPipelineCreator
+from wb.main.shared.enumerates import TaskEnum
 from wb.main.utils.utils import create_empty_dir
 
 
@@ -59,7 +62,12 @@ def __init__(self, model_id: int, huggingface_model_id: str, model_name: str):
     def create_result_model(self, session: Session):
         uploaded_model: TopologiesModel = session.query(TopologiesModel).get(self.model_id)
 
+        # consider all huggingface models as text classification
         metadata = TopologiesMetaDataModel()
+        metadata.task_type = TaskEnum.text_classification.value
+        accuracy_config = json.loads(DEFAULT_ACCURACY_CONFIGURATION)
+        accuracy_config['taskType'] = TaskEnum.text_classification.value
+        metadata.advanced_configuration = json.dumps(accuracy_config)
         metadata.write_record(session)
 
         self._result_model = TopologiesModel(self.model_name, SupportedFrameworksEnum.openvino, metadata.id)