add new models and fix sessions

aana_chat_with_video/alembic/versions/b9860676dd49_set_server_default_for_task_completed_.py

@@ -0,0 +1,47 @@
+"""Set server default for task.completed_at and task.assigned_at to none and add num_retries.
+Revision ID: b9860676dd49
+Revises: 5ad873484aa3
+Create Date: 2024-08-22 07:54:55.921710
+"""
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "b9860676dd49"
+down_revision: str | None = "5ad873484aa3"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+
+def upgrade() -> None:
+    """Upgrade database to this revision from previous."""
+    with op.batch_alter_table("tasks", schema=None) as batch_op:
+        batch_op.alter_column(
+            "completed_at",
+            server_default=None,
+        )
+        batch_op.alter_column(
+            "assigned_at",
+            server_default=None,
+        )
+        batch_op.add_column(
+            sa.Column(
+                "num_retries",
+                sa.Integer(),
+                nullable=False,
+                comment="Number of retries",
+                server_default=sa.text("0"),
+            )
+        )
+
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade database from this revision to previous."""
+    with op.batch_alter_table("tasks", schema=None) as batch_op:
+        batch_op.drop_column("num_retries")
+
+    # ### end Alembic commands ###

aana_chat_with_video/alembic/versions/d93a90261ee5_added_extended_video.py

@@ -1,7 +1,7 @@
 """added extended video.
 
 Revision ID: d93a90261ee5
-Revises: 5ad873484aa3
+Revises: b9860676dd49
 Create Date: 2024-07-29 12:41:04.976640
 
 """
@@ -12,7 +12,7 @@
 
 # revision identifiers, used by Alembic.
 revision: str = 'd93a90261ee5'
-down_revision: str | None = '5ad873484aa3'
+down_revision: str | None = 'b9860676dd49'
 branch_labels: str | Sequence[str] | None = None
 depends_on: str | Sequence[str] | None = None
 

aana_chat_with_video/configs/deployments.py

@@ -1,6 +1,5 @@
 from aana.core.models.sampling import SamplingParams
 from aana.core.models.types import Dtype
-from aana.deployments.hf_blip2_deployment import HFBlip2Config, HFBlip2Deployment
 from aana.deployments.vad_deployment import VadConfig, VadDeployment
 from aana.deployments.vllm_deployment import VLLMConfig, VLLMDeployment
 from aana.deployments.whisper_deployment import (
@@ -18,7 +17,7 @@
             max_ongoing_requests=1000,
             ray_actor_options={"num_gpus": 0.25},
             user_config=WhisperConfig(
-                model_size=WhisperModelSize.MEDIUM,
+                model_size=WhisperModelSize.TURBO,
                 compute_type=WhisperComputeType.FLOAT16,
             ).model_dump(mode="json"),
         ),
@@ -41,15 +40,19 @@
     },
     {
         "name": "captioning_deployment",
-        "instance": HFBlip2Deployment.options(
+        "instance": VLLMDeployment.options(
             num_replicas=1,
-            max_ongoing_requests=1000,
             ray_actor_options={"num_gpus": 0.25},
-            user_config=HFBlip2Config(
-                model="Salesforce/blip2-opt-2.7b",
-                dtype=Dtype.FLOAT16,
-                batch_size=2,
-                num_processing_threads=2,
+            user_config=VLLMConfig(
+                model="Qwen/Qwen2-VL-2B-Instruct",
+                dtype=Dtype.AUTO,
+                gpu_memory_reserved=12000,
+                max_model_len=32768,
+                enforce_eager=True,
+                default_sampling_params=SamplingParams(
+                    temperature=0.0, top_p=1.0, top_k=-1, max_tokens=512
+                ),
+                engine_args={"trust_remote_code": True},
             ).model_dump(mode="json"),
         ),
     },

aana_chat_with_video/configs/settings.py

@@ -5,7 +5,7 @@ class Settings(AanaSettings):
     """A pydantic model for App settings."""
 
     asr_model_name: str = "whisper_medium"
-    captioning_model_name: str = "hf_blip2_opt_2_7b"
+    captioning_model_name: str = "qwen2-vl-2b-instruct"
     max_video_len: int = 60 * 20  # 20 minutes
 
 

aana_chat_with_video/endpoints/delete_video.py

@@ -2,6 +2,7 @@
 
 from aana.api.api_generation import Endpoint
 from aana.core.models.media import MediaId
+from aana.storage.session import get_session
 from aana_chat_with_video.storage.repository.extended_video import (
     ExtendedVideoRepository,
 )
@@ -19,9 +20,9 @@ class DeleteVideoEndpoint(Endpoint):
     async def initialize(self):
         """Initialize the endpoint."""
         await super().initialize()
-        self.video_repo = ExtendedVideoRepository(self.session)
 
     async def run(self, media_id: MediaId) -> DeleteVideoOutput:
         """Delete video."""
-        self.video_repo.delete(media_id)
+        with get_session() as session:
+            ExtendedVideoRepository(session).delete(media_id)
         return DeleteVideoOutput(media_id=media_id)