feat(labeler): gently terminate the labler UI from frontend

finetuner/__init__.py

@@ -26,7 +26,7 @@ def fit(
     optimizer: str = 'adam',
     optimizer_kwargs: Optional[Dict] = None,
     device: str = 'cpu',
-) -> 'Summary':
+) -> Tuple['AnyDNN', 'Summary']:
     ...
 
 
@@ -49,7 +49,7 @@ def fit(
     output_dim: Optional[int] = None,
     freeze: bool = False,
     device: str = 'cpu',
-) -> 'Summary':
+) -> Tuple['AnyDNN', 'Summary']:
     ...
 
 
@@ -67,7 +67,7 @@ def fit(
     optimizer: str = 'adam',
     optimizer_kwargs: Optional[Dict] = None,
     device: str = 'cpu',
-) -> None:
+) -> Tuple['AnyDNN', 'Summary']:
     ...
 
 
@@ -91,13 +91,13 @@ def fit(
     output_dim: Optional[int] = None,
     freeze: bool = False,
     device: str = 'cpu',
-) -> None:
+) -> Tuple['AnyDNN', 'Summary']:
     ...
 
 
 def fit(
     model: 'AnyDNN', train_data: 'DocumentArrayLike', *args, **kwargs
-) -> Optional[Tuple['AnyDNN', 'Summary']]:
+) -> Tuple['AnyDNN', Optional['Summary']]:
     if kwargs.get('to_embedding_model', False):
         from .tailor import to_embedding_model
 
@@ -106,8 +106,6 @@ def fit(
     if kwargs.get('interactive', False):
         from .labeler import fit
 
-        # TODO: atm return will never hit as labeler UI hangs the
-        #  flow via `.block()`
         return model, fit(model, train_data, *args, **kwargs)
     else:
         from .tuner import fit

finetuner/labeler/__init__.py

@@ -1,5 +1,6 @@
 import os
 import tempfile
+import threading
 import webbrowser
 from typing import Optional
 
@@ -36,11 +37,15 @@ def fit(
     :param kwargs: Additional keyword arguments.
     """
     dam_path = tempfile.mkdtemp()
+    stop_event = threading.Event()
 
     class MyExecutor(FTExecutor):
         def get_embed_model(self):
             return embed_model
 
+        def get_stop_event(self):
+            return stop_event
+
     f = (
         Flow(
             protocol='http',
@@ -70,6 +75,7 @@ def get_embed_model(self):
     f.expose_endpoint(
         '/save'
     )  #: for signaling the backend to save the current state of the model
+    f.expose_endpoint('/terminate')  #: for terminating the flow from frontend
 
     def extend_rest_function(app):
         """Allow FastAPI frontend to serve finetuner UI as a static webpage"""
@@ -107,4 +113,4 @@ def open_frontend_in_browser(req):
             show_progress=True,
             on_done=open_frontend_in_browser,
         )
-        f.block()
+        f.block(stop_event=stop_event)

finetuner/labeler/executor.py

@@ -25,6 +25,10 @@ def __init__(
     def get_embed_model(self):
         ...
 
+    @abc.abstractmethod
+    def get_stop_event(self):
+        ...
+
     @cached_property
     def _embed_model(self):
         return self.get_embed_model()
@@ -65,6 +69,10 @@ def save(self, parameters: Dict, **kwargs):
         save(self._embed_model, model_path)
         print(f'model is saved to {model_path}')
 
+    @requests(on='/terminate')
+    def terminate(self, **kwargs):
+        self.get_stop_event().set()
+
 
 class DataIterator(Executor):
     def __init__(

finetuner/labeler/ui/index.html

@@ -22,7 +22,7 @@
         </div>
         <sidebar :labeler-config="labeler_config" :view-template="view_template" :tags="tags" :is-busy="is_busy"
             :progress-stats="progress_stats" :positive-rate="positive_rate" :negative-rate="negative_rate"
-            :advanced-config="advanced_config" :save-progress="saveProgress"
+            :advanced-config="advanced_config" :save-progress="saveProgress" :terminate-flow="terminateFlow"
                  :next-batch="next_batch"></sidebar>
         <div class="b-example-divider"></div>
         <div class="flex-grow-1 p-1 overflow-hidden">

finetuner/labeler/ui/js/components/sidebar.vue.js

@@ -10,6 +10,7 @@ const sidebar = {
         advancedConfig: Object,
         saveProgress: Function,
         nextBatch: Function,
+        terminateFlow: Function,
     },
     template: `
     <div class="d-flex flex-column flex-shrink-0 p-3 sidebar">
@@ -124,10 +125,14 @@ const sidebar = {
                         </div>
                     </div>
                     <div class="my-3 d-flex justify-content-center">
-                      <button class="btn btn btn-outline-primary"
+                      <button class="btn btn btn-outline-primary m-2"
                           v-on:click="saveProgress()">
                           Save model
                       </button>
+                      <button class="btn btn btn-outline-secondary m-2"
+                          v-on:click="terminateFlow()">
+                          Terminate
+                      </button>
                     </div>
                 </div>
             </div>

finetuner/labeler/ui/js/main.js

@@ -41,7 +41,8 @@ const app = new Vue({
             server_address: `http://localhost`,
             next_endpoint: '/next',
             fit_endpoint: '/fit',
-            saveEndpoint: '/save',
+            save_endpoint: '/save',
+            stop_endpoint: '/terminate',
         },
         advanced_config: {
             pos_value: {text: 'Positive label', value: 1, type: 'number'},
@@ -80,8 +81,11 @@ const app = new Vue({
         fit_address: function () {
             return `${this.host_address}${this.general_config.fit_endpoint}`
         },
-        saveAddress: function () {
-            return `${this.host_address}${this.general_config.saveEndpoint}`
+        save_address: function () {
+            return `${this.host_address}${this.general_config.save_endpoint}`
+        },
+        stop_address: function () {
+            return `${this.host_address}${this.general_config.stop_endpoint}`
         },
         positive_rate: function () {
             return this.progress_stats.positive.value / (this.progress_stats.positive.value + this.progress_stats.negative.value) * 100
@@ -208,7 +212,7 @@ const app = new Vue({
             app.is_conn_broken = false
             $.ajax({
                 type: "POST",
-                url: app.saveAddress,
+                url: app.save_address,
                 data: JSON.stringify({
                     data: [],
                     parameters: {
@@ -225,6 +229,27 @@ const app = new Vue({
                 app.is_busy = false
             });
         },
+        terminateFlow: () => {
+            app.is_busy = true
+            app.is_conn_broken = false
+            $.ajax({
+                type: "POST",
+                url: app.stop_address,
+                data: JSON.stringify({
+                    data: [],
+                    parameters: {
+                    }
+                }),
+                contentType: "application/json; charset=utf-8",
+                dataType: "json",
+            }).success(function (data, textStatus, jqXHR) {
+                app.is_busy = false
+                close();
+            }).fail(function () {
+                console.error("Error: ", error)
+                app.is_busy = false
+            });
+        },
         handleKeyPress(event) {
             let key = event.key
             if (event.target instanceof HTMLInputElement) {

finetuner/tuner/summary.py

@@ -48,6 +48,11 @@ def __init__(self, *records: ScalarSequence):
         """Create a collection of summaries. """
         self._records = [r for r in records if r]
 
+    def __iadd__(self, other: 'Summary'):
+        if isinstance(other, Summary):
+            self._records += other._records
+        return self
+
     def save(self, filepath: str):
         """Store all summary into a JSON file"""
         with open(filepath, 'w') as fp:

tests/integration/labeler/test_tune_lstm.py

@@ -55,14 +55,17 @@ def _run(framework_name, loss, port_expose):
         ),
     }
 
-    fit(
+    rv1, rv2 = fit(
         embed_models[framework_name](),
         generate_qa_match(num_total=10, num_neg=0),
         loss=loss,
         interactive=True,
         port_expose=port_expose,
     )
 
+    assert rv1
+    assert not rv2
+
 
 all_test_losses = [
     'CosineSiameseLoss',
@@ -144,6 +147,16 @@ def test_all_frameworks(framework, loss, tmpdir):
         assert req.status_code == 200
         assert os.path.isfile(model_path)
 
+        req = requests.post(
+            f'http://localhost:{port}/terminate',
+            json={
+                'data': [],
+                'parameters': {},
+            },
+        )
+        assert req.status_code == 200
+        assert os.path.isfile(model_path)
+
     except:
         raise
     finally:

tests/integration/labeler/test_tune_mlp.py

@@ -53,21 +53,24 @@ def _run(framework_name, loss, port_expose):
         ),
     }
 
-    fit(
+    rv1, rv2 = fit(
         embed_models[framework_name](),
         generate_fashion_match(num_total=10, num_pos=0, num_neg=0),
         loss=loss,
         interactive=True,
         port_expose=port_expose,
     )
 
+    assert rv1
+    assert not rv2
+
 
 # 'keras' does not work under this test setup
 # Exception ... ust be from the same graph as Tensor ...
 # TODO: add keras backend back to the test
 @pytest.mark.parametrize('framework', ['pytorch', 'paddle'])
 @pytest.mark.parametrize('loss', all_test_losses)
-def test_all_frameworks(framework, loss):
+def test_all_frameworks(framework, loss, tmpdir):
     port = random_port()
     p = multiprocessing.Process(
         target=_run,
@@ -125,6 +128,27 @@ def test_all_frameworks(framework, loss):
             json={'data': rj['data']['docs'], 'parameters': {'epochs': 10}},
         )
         assert req.status_code == 200
+
+        model_path = os.path.join(tmpdir, 'model.train')
+        req = requests.post(
+            f'http://localhost:{port}/save',
+            json={
+                'data': [],
+                'parameters': {'model_path': model_path},
+            },
+        )
+        assert req.status_code == 200
+        assert os.path.isfile(model_path)
+
+        req = requests.post(
+            f'http://localhost:{port}/terminate',
+            json={
+                'data': [],
+                'parameters': {},
+            },
+        )
+        assert req.status_code == 200
+
     except:
         raise
     finally: