Fix SQL multimodal example

examples/sql-example.ipynb

@@ -62,6 +62,16 @@
     "The initial step in any `superduperdb` workflow is to connect to your datastore. To connect to a different datastore, simply add a different `URI`, for example, `postgres://...`."
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "587bffd3-7202-4e12-a6de-435d4b16892d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!rm .superduperdb/test.ddb"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -78,6 +88,16 @@
     "db = superduper('duckdb://.superduperdb/test.ddb')"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "0a5252a3-91a7-4789-85b7-0b8279c228a3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "db.show('vector_index')"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "b8794451",
@@ -108,7 +128,7 @@
     "!mkdir -p data/coco\n",
     "\n",
     "# Move the 'images_small' directory to 'data/coco/images'\n",
-    "!mv images_small data/coco/images"
+    "!mv images_tiny data/coco/images"
    ]
   },
   {
@@ -224,7 +244,7 @@
    "source": [
     "## Build SuperDuperDB `Model` Instances\n",
     "\n",
-    "This use-case utilizes the `superduperdb.ext.torch` extension. Both models used output `torch` tensors, which are encoded with `tensor`:"
+    "This use-case utilizes the `superduperdb.ext.torch` extension. Both models use `torch` tensors in their output, which are encoded with `tensor`:"
    ]
   },
   {

examples/transfer_learning.ipynb

@@ -4,7 +4,10 @@
    "cell_type": "markdown",
    "id": "fe6fd0ab0e1ad844",
    "metadata": {
-    "collapsed": false
+    "collapsed": false,
+    "jupyter": {
+     "outputs_hidden": false
+    }
    },
    "source": [
     "# Transfer Learning with Sentence Transformers and Scikit-Learn"
@@ -14,7 +17,10 @@
    "cell_type": "markdown",
    "id": "8dcde44d942793ff",
    "metadata": {
-    "collapsed": false
+    "collapsed": false,
+    "jupyter": {
+     "outputs_hidden": false
+    }
    },
    "source": [
     "## Introduction\n",
@@ -26,7 +32,10 @@
    "cell_type": "markdown",
    "id": "1809feca8a8dca5a",
    "metadata": {
-    "collapsed": false
+    "collapsed": false,
+    "jupyter": {
+     "outputs_hidden": false
+    }
    },
    "source": [
     "## Prerequisites\n",
@@ -41,7 +50,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "!pip install superduperdb\n",
+    "# !pip install superduperdb\n",
     "!pip install ipython numpy datasets sentence-transformers"
    ]
   },
@@ -57,7 +66,10 @@
    "cell_type": "markdown",
    "id": "5379007991707d17",
    "metadata": {
-    "collapsed": false
+    "collapsed": false,
+    "jupyter": {
+     "outputs_hidden": false
+    }
    },
    "source": [
     "First, we need to establish a connection to a MongoDB datastore via SuperDuperDB. You can configure the `MongoDB_URI` based on your specific setup. \n",
@@ -84,7 +96,7 @@
     "\n",
     "# SuperDuperDB, now handles your MongoDB database\n",
     "# It just super dupers your database\n",
-    "db = superduper(mongodb_uri)\n",
+    "db = superduper(mongodb_uri, artifact_store='filesystem://./data/')\n",
     "\n",
     "# Reference a collection called transfer\n",
     "collection = Collection('transfer')"
@@ -169,10 +181,21 @@
     "    X='text',\n",
     "    db=db,\n",
     "    select=collection.find(),\n",
-    "    listen=True\n",
+    "    listen=True,\n",
+    "    show_progress_bar=True,\n",
     ")"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "aed95196-4454-4d88-beb8-dc6dddfcbe33",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "db.execute(collection.find_one())"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "68fefc17",
@@ -201,10 +224,10 @@
     "\n",
     "# Train the model on 'text' data with corresponding labels\n",
     "model.fit(\n",
-    "    X='text',\n",
+    "    X='_outputs.text.all-MiniLM-L6-v2.0',\n",
     "    y='label',\n",
     "    db=db,\n",
-    "    select=collection.find().featurize({'text': 'all-MiniLM-L6-v2'}),\n",
+    "    select=collection.find(),\n",
     ")\n"
    ]
   },
@@ -227,9 +250,9 @@
    "source": [
     "# Make predictions on 'text' data with the trained SuperDuperDB model\n",
     "model.predict(\n",
-    "    X='text',\n",
+    "    X='_outputs.text.all-MiniLM-L6-v2.0',\n",
     "    db=db,\n",
-    "    select=collection.find().featurize({'text': 'all-MiniLM-L6-v2'}),\n",
+    "    select=collection.find(),\n",
     "    listen=True,\n",
     ")"
    ]
@@ -258,16 +281,8 @@
     "print(r['text'][:100])\n",
     "\n",
     "# Print the prediction made by the SVC model stored in '_outputs'\n",
-    "print(r['_outputs']['text']['svc'])"
+    "print(r['_outputs']['text']['svc']['0'])"
    ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "2174a0a3-c32e-4481-a301-370b569ba30c",
-   "metadata": {},
-   "outputs": [],
-   "source": []
   }
  ],
  "metadata": {
@@ -286,7 +301,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.5"
+   "version": "3.11.6"
   }
  },
  "nbformat": 4,

examples/voice_memos.ipynb

@@ -158,15 +158,15 @@
     "db.execute(voice_collection.insert_many([\n",
     "    # Create a SuperDuperDB Document for each audio sample\n",
     "    Document({'audio': enc(r['audio']['array'])}) for r in data\n",
-    "]))\n"
+    "]))"
    ]
   },
   {
    "cell_type": "markdown",
    "id": "721f31f4626881e0",
    "metadata": {},
    "source": [
-    "## Install Pre-Trained Model (LibriSpeech) into Database\n",
+    "## Install Pre-Trained Model (LibriSpeech) with Database\n",
     "\n",
     "Apply a pre-trained `transformers` model to the data:"
    ]
@@ -225,7 +225,7 @@
     "    db=db,  # Provide the SuperDuperDB instance\n",
     "    select=voice_collection.find(),  # Specify the collection of audio data to transcribe\n",
     "    max_chunk_size=10  # Set the maximum chunk size for processing audio data\n",
-    ")\n"
+    ")"
    ]
   },
   {
@@ -256,7 +256,7 @@
     "        identifier='my-index',  # Set a unique identifier for the VectorIndex\n",
     "        indexing_listener=Listener(\n",
     "            model=OpenAIEmbedding(model='text-embedding-ada-002'),  # Use OpenAIEmbedding for audio transcriptions\n",
-    "            key='_outputs.audio.transcription',  # Specify the key for indexing the transcriptions in the output\n",
+    "            key='_outputs.audio.transcription.0',  # Specify the key for indexing the transcriptions in the output\n",
     "            select=voice_collection.find(),  # Select the collection of audio data to index\n",
     "        ),\n",
     "    )\n",
@@ -290,7 +290,7 @@
     "search_results = list(\n",
     "    db.execute(\n",
     "        voice_collection.like(\n",
-    "            {'_outputs.audio.transcription': search_term},\n",
+    "            {'_outputs.audio.transcription.0': search_term},\n",
     "            n=num_results,\n",
     "            vector_index='my-index',  # Use the 'my-index' VectorIndex for similarity search\n",
     "        ).find({}, {'_outputs.audio.transcription': 1})  # Retrieve only the 'transcription' field in the results\n",
@@ -370,11 +370,11 @@
     "    \n",
     "    # Select relevant context for the model from the SuperDuperDB collection of audio transcriptions\n",
     "    context_select=voice_collection.like(\n",
-    "        Document({'_outputs.audio.transcription': question}), vector_index='my-index'\n",
+    "        Document({'_outputs.audio.transcription.0': question}), vector_index='my-index'\n",
     "    ).find(),\n",
     "    \n",
     "    # Specify the key in the context used by the model\n",
-    "    context_key='_outputs.audio.transcription',\n",
+    "    context_key='_outputs.audio.transcription.0',\n",
     ")[0].content\n",
     "\n",
     "# Print the response obtained from the chat completion model\n",
@@ -398,7 +398,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.5"
+   "version": "3.11.6"
   }
  },
  "nbformat": 4,

superduperdb/base/datalayer.py

@@ -168,7 +168,11 @@ def backfill_vector_search(self, vi, searcher):
 
                 id = record[self.databackend.id_field]
                 assert not isinstance(vi.indexing_listener.model, str)
-                h = record.outputs(key, vi.indexing_listener.model.identifier)
+                h = record.outputs(
+                    key,
+                    vi.indexing_listener.model.identifier,
+                    version=vi.indexing_listener.model.version,
+                )
                 if isinstance(h, Encodable):
                     h = h.x
                 items.append(VectorItem.create(id=str(id), vector=h))
@@ -915,7 +919,6 @@ def _add(
             object.on_load(self)
             return object.schedule_jobs(self, dependencies=dependencies), object
         except Exception as e:
-
             raise exceptions.DatalayerException(
                 f'Error while adding object with id: {object.identifier}'
             ) from e

superduperdb/base/document.py

@@ -7,6 +7,7 @@
 from superduperdb.components.encoder import Encodable, Encoder
 from superduperdb.components.schema import Schema
 from superduperdb.misc.files import get_file_from_uri
+from superduperdb.misc.special_dicts import MongoStyleDict
 
 ContentType = t.Union[t.Dict, Encodable]
 ItemType = t.Union[t.Dict[str, t.Any], Encodable, ObjectId]
@@ -46,10 +47,11 @@ def outputs(self, key: str, model: str, version: t.Optional[int] = None) -> t.An
         :param key: Document key to get outputs from.
         :param model: Model name to get outputs from.
         """
+        r = MongoStyleDict(self.unpack())
         if version is not None:
-            document = self.unpack()[_OUTPUTS_KEY][key][model][version]
+            document = r[f'{_OUTPUTS_KEY}.{key}.{model}.{version}']
         else:
-            tmp = self.unpack()[_OUTPUTS_KEY][key][model]
+            tmp = r[f'{_OUTPUTS_KEY}.{key}.{model}']
             version = max(list(tmp.keys()))
             return tmp[version]
         return document

superduperdb/components/model.py

@@ -34,6 +34,15 @@
 ObjectsArg = t.Sequence[t.Union[t.Any, Artifact]]
 
 
+class _to_call:
+    def __init__(self, callable, **kwargs):
+        self.callable = callable
+        self.kwargs = kwargs
+
+    def __call__(self, X):
+        return self.callable(X, **self.kwargs)
+
+
 @dc.dataclass
 class _TrainingConfiguration(Component):
     """
@@ -133,20 +142,23 @@ def _predict_one(self, X: t.Any, **kwargs) -> int:
 
         return output
 
-    def _forward(self, X: t.Sequence[int], num_workers: int = 0) -> t.Sequence[int]:
+    def _forward(
+        self, X: t.Sequence[int], num_workers: int = 0, **kwargs
+    ) -> t.Sequence[int]:
         if self.batch_predict:
-            return self.to_call(X)
+            return self.to_call(X, **kwargs)
 
         outputs = []
         if num_workers:
+            to_call = _to_call(self.to_call, **kwargs)
             pool = multiprocessing.Pool(processes=num_workers)
-            for r in pool.map(self.to_call, X):
+            for r in pool.map(to_call, X):
                 outputs.append(r)
             pool.close()
             pool.join()
         else:
             for r in X:
-                outputs.append(self.to_call(r))
+                outputs.append(self.to_call(r, **kwargs))
         return outputs
 
     def _predict(

superduperdb/misc/download.py

@@ -379,6 +379,7 @@ def download_content(
             pass
 
     if CFG.hybrid_storage:
+        assert isinstance(CFG.downloads_folder, str)
         _download_update = SaveFile(CFG.downloads_folder)
     else:
 

superduperdb/misc/special_dicts.py

@@ -40,6 +40,9 @@ def __setitem__(self, key: str, value: t.Any) -> None:
         else:
             parent = key.split('.')[0]
             child = '.'.join(key.split('.')[1:])
-            parent_item = MongoStyleDict(self[parent])
+            try:
+                parent_item = MongoStyleDict(self[parent])
+            except KeyError:
+                parent_item = MongoStyleDict({})
             parent_item[child] = value
             self[parent] = parent_item

test/unittest/misc/test_downloaders.py

@@ -20,7 +20,6 @@ def test_s3_and_web():
 
 @pytest.fixture
 def patch_cfg_downloads(monkeypatch):
-    monkeypatch.setattr(CFG, 'hybrid_storage', True)
     td = str(uuid.uuid4())
     with tempfile.TemporaryDirectory() as td:
         monkeypatch.setattr(CFG, 'downloads_folder', td)