Merge branch 'master' into new-branch-8

Author: vhaldemar

examples/async/completions/batch.py

@@ -61,19 +61,19 @@ async def main() -> None:
 
     # There are only specific models supports batch runs,
     # please refer to Yandex Foundation Models doctumentation
-    model = sdk.models.completions('gemma-3-12b-it')
+    model = sdk.models.completions('qwen2.5-32b-instruct')
 
     # Batch run returns an operation object you could
     # follow or just call .wait method
     task = await model.batch.run_deferred(dataset)
 
-    try:
-        resulting_dataset = await task
-    except:
-        status = await task.get_status()
-        if status.is_running:
-            await task.cancel()
-        raise
+    # You also could use "structured output feature".
+    # Look for details in examples/async/completions/structured_output.py
+    model = model.configure(response_format='json')
+    json_task = await model.batch.run_deferred(dataset)
+
+    resulting_dataset = await task
+    json_dataset = await json_task
 
     # NB: Resulting dataset have task_type="TextToTextGeneration" and could be used as an input
     # for tuning.
@@ -84,11 +84,18 @@ async def main() -> None:
         print('Resulting dataset lines:')
         async for line in resulting_dataset.read():
             print(line)
+
+        print('Json dataset lines:')
+        async for line in json_dataset.read():
+            print(line)
     except ImportError:
         print('skipping dataset read; install yandex-cloud-ml-sdk[datasets] to be able to read')
 
-    # Removing final dataset to not to increase chaos.
+    # Removing all the data to not to increase chaos.
     await resulting_dataset.delete()
+    await json_dataset.delete()
+    await task.delete()
+    await json_task.delete()
 
 
 if __name__ == '__main__':

examples/sync/completions/batch.py

@@ -60,13 +60,19 @@ def main() -> None:
 
     # There are only specific models supports batch runs,
     # please refer to Yandex Foundation Models doctumentation
-    model = sdk.models.completions('gemma-3-12b-it')
+    model = sdk.models.completions('qwen2.5-32b-instruct')
 
     # Batch run returns an operation object you could
     # follow or just call .wait method
-    operation = model.batch.run_deferred(dataset)
+    task = model.batch.run_deferred(dataset)
 
-    resulting_dataset = operation.wait()
+    # You also could use "structured output feature".
+    # Look for details in examples/sync/completions/structured_output.py
+    model = model.configure(response_format='json')
+    json_task = model.batch.run_deferred(dataset)
+
+    resulting_dataset = task.wait()
+    json_dataset = json_task.wait()
 
     # NB: Resulting dataset have task_type="TextToTextGeneration" and could be used as an input
     # for tuning.
@@ -77,11 +83,19 @@ def main() -> None:
         print('Resulting dataset lines:')
         for line in resulting_dataset.read():
             print(line)
+
+        print('Json dataset lines:')
+        for line in json_dataset.read():
+            print(line)
     except ImportError:
-        print('skipping dataset read; istall yandex-cloud-ml-sdk[datasets] to be able to read')
+        print('skipping dataset read; install yandex-cloud-ml-sdk[datasets] to be able to read')
 
-    # Removing final dataset to not to increase chaos.
+    # Removing all the data to not to increase chaos.
     resulting_dataset.delete()
+    json_dataset.delete()
+    task.delete()
+    json_task.delete()
+
 
 if __name__ == '__main__':
     main()

pyproject.toml

@@ -33,7 +33,7 @@ classifiers = [
 requires-python = ">=3.9"
 dynamic = ["version"]
 dependencies = [
-    "yandexcloud>=0.350.0",
+    "yandexcloud>=0.354.0",
     "grpcio>=1.70.0",
     "get-annotations",
     "httpx>=0.27,<1",

src/yandex_cloud_ml_sdk/_models/completions/model.py

@@ -183,7 +183,7 @@ def _make_request(
         )
 
     def _make_batch_request(self, dataset_id: str) -> BatchCompletionRequest:
-        for field in ('tools', 'response_format', 'tool_choice', 'parallel_tool_calls'):
+        for field in ('tools', 'tool_choice', 'parallel_tool_calls'):
             value = getattr(self.config, field)
             if value is not None:
                 warnings.warn(
@@ -192,10 +192,13 @@ def _make_batch_request(self, dataset_id: str) -> BatchCompletionRequest:
                     UserWarning, 4
                 )
 
+        response_format_kwargs = make_response_format_kwargs(self._config.response_format)
+
         return BatchCompletionRequest(
             model_uri=self.uri,
             completion_options=self._make_completion_options(stream=False),
-            source_dataset_id=dataset_id
+            source_dataset_id=dataset_id,
+            **response_format_kwargs,
         )
 
     async def _run_sync_impl(

src/yandex_cloud_ml_sdk/_types/batch/domain.py

@@ -40,7 +40,7 @@ async def _run_deferred(self, dataset: DatasetType, *, timeout: float = 60) -> B
         stub_class = m._batch_service_stub
         proto_metadata_type = m._batch_proto_metadata_type
 
-        logger.debug(
+        logger.info(
             'going to create batch task at %r service with a %r request',
             stub_class.__name__, request
         )