Add UI for dataset settings to edit tags of a dataset. (databricks#824)

- Move tags up to the top-level dataset settings in python
- Allow the UI to change the tags.
- When some datasets are tagged, show a no-tag for the untagged ones. If
no datasets are tagged, no tag UI is shown.

Demo: https://huggingface.co/spaces/lilacai/nikhil_staging


https://github.com/lilacai/lilac/assets/1100749/d0077f40-fd2b-4d24-b983-e299306a2e7c

Other fixes:
- Move the skeleton text for the row_id query to the bottom so it's not
always at the top.
- Fix some padding in the settings modal that caused focus to be cut
off.

lilac/config.py

@@ -121,15 +121,22 @@ class DatasetSettings(BaseModel):
   ui: Optional[DatasetUISettings] = None
   preferred_embedding: Optional[str] = DEFAULT_EMBEDDING
   model_config = ConfigDict(extra='forbid')
+  tags: Optional[list[str]] = PydanticField(
+    description='A list of tags for the dataset to organize in the UI.', default=[]
+  )
 
 
 class DatasetConfig(BaseModel):
   """Configures a dataset with a source and transformations."""
 
   namespace: str = PydanticField(description='The namespace of the dataset.')
   name: str = PydanticField(description='The name of the dataset.')
-  tags: list[str] = PydanticField(
-    description='A list of tags for the dataset to organize in the UI.', default=[]
+
+  # Deprecated.
+  tags: Optional[list[str]] = PydanticField(
+    description='[Deprecated] This field is *deprecated* in favor of DatasetSettings.tags and '
+    'will be removed in a later release.',
+    default=[],
   )
 
   # The source configuration.

lilac/db_manager.py

@@ -97,9 +97,9 @@ def list_datasets(project_dir: Optional[Union[str, pathlib.Path]] = None) -> lis
         continue
 
       dataset_config = get_dataset_config(project_config, namespace, dataset_name)
-      tags = []
-      if dataset_config:
-        tags = dataset_config.tags
+      tags: list[str] = []
+      if dataset_config and dataset_config.settings and dataset_config.settings.tags:
+        tags = dataset_config.settings.tags
 
       dataset_infos.append(DatasetInfo(namespace=namespace, dataset_name=dataset_name, tags=tags))
 

lilac_hf_space.yml

@@ -1,8 +1,8 @@
 datasets:
   - name: imdb
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - text
@@ -13,8 +13,8 @@ datasets:
 
   - name: open-asssistant-conversations
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - text
@@ -25,8 +25,8 @@ datasets:
 
   - name: wikitext-2-raw-v1
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - text
@@ -39,11 +39,11 @@ datasets:
 
   - name: textbook_quality_programming
     namespace: lilac
-    tags: [machine-learning]
     source:
       dataset_name: vikp/textbook_quality_programming
       source_name: huggingface
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - - outline
@@ -57,8 +57,8 @@ datasets:
 
   - name: databricks-dolly-15k-curated-en
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - original-instruction
@@ -80,8 +80,8 @@ datasets:
 
   - name: 'OpenOrca-100k'
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - question
@@ -94,8 +94,8 @@ datasets:
 
   - name: opus100-en-es-validation
     namespace: lilac
-    tags: [machine-learning]
     settings:
+      tags: [machine-learning]
       ui:
         media_paths:
           - [translation, es]
@@ -110,8 +110,8 @@ datasets:
   # Science datasets
   - name: science-qa-derek-thomas
     namespace: lilac
-    tags: [science]
     settings:
+      tags: [science]
       ui:
         media_paths: [lecture]
       preferred_embedding: 'gte-small'
@@ -122,8 +122,8 @@ datasets:
   # Business datasets.
   - name: enron-emails
     namespace: lilac
-    tags: [business]
     settings:
+      tags: [business]
       ui:
         media_paths: [text]
       preferred_embedding: 'gte-small'
@@ -136,8 +136,8 @@ datasets:
   # Other datasets.
   - name: the_movies_dataset
     namespace: lilac
-    tags: [other]
     settings:
+      tags: [other]
       ui:
         media_paths: [overview]
       preferred_embedding: 'gte-small'

scripts/deploy_staging.py

@@ -18,6 +18,7 @@
     files.
   --skip_data_upload: When true, only uploads the wheel + ts files without any other changes.
   --create_space: When true, creates the space if it doesn't exist.
+
 """
 
 import os
@@ -27,7 +28,6 @@
 
 import click
 from huggingface_hub import CommitOperationAdd, CommitOperationDelete, HfApi
-
 from lilac.deploy import PY_DIST_DIR, deploy_project_operations
 from lilac.env import env
 from lilac.utils import log
@@ -38,51 +38,60 @@
   '--hf_space',
   help='The huggingface space. Defaults to env.HF_STAGING_DEMO_REPO. '
   'Should be formatted like `SPACE_ORG/SPACE_NAME`.',
-  type=str)
+  type=str,
+)
 @click.option('--dataset', help='The name of a dataset to upload', type=str, multiple=True)
 @click.option(
   '--concept',
   help='The name of a concept to upload. By default all lilac/ concepts are uploaded.',
   type=str,
-  multiple=True)
+  multiple=True,
+)
 @click.option(
   '--skip_ts_build',
   help='Skip building the web server TypeScript. '
   'Useful to speed up the build if you are only changing python or data.',
   type=bool,
   is_flag=True,
-  default=False)
+  default=False,
+)
 @click.option(
   '--skip_cache_upload',
   help='Skip uploading the cache files from .cache/lilac which contain cached concept pkl models.',
   type=bool,
   is_flag=True,
-  default=False)
+  default=False,
+)
 @click.option(
   '--skip_ts_build',
   help='Skip building the web server TypeScript. '
   'Useful to speed up the build if you are only changing python or data.',
   type=bool,
   is_flag=True,
-  default=False)
+  default=False,
+)
 @click.option(
   '--create_space',
   help='When True, creates the HuggingFace space if it doesnt exist. The space will be created '
   'with the storage type defined by --hf_space_storage.',
   is_flag=True,
-  default=False)
+  default=False,
+)
 @click.option(
   '--skip_data_upload',
   help='When true, only uploads the wheel files without any other changes.',
   is_flag=True,
-  default=False)
-def deploy_staging(hf_space: Optional[str] = None,
-                   dataset: Optional[list[str]] = None,
-                   concept: Optional[list[str]] = None,
-                   skip_cache_upload: Optional[bool] = False,
-                   skip_ts_build: Optional[bool] = False,
-                   skip_data_upload: Optional[bool] = False,
-                   create_space: Optional[bool] = False) -> None:
+  default=False,
+)
+def deploy_staging(
+  hf_space: Optional[str] = None,
+  dataset: Optional[list[str]] = None,
+  concept: Optional[list[str]] = None,
+  skip_cache_upload: Optional[bool] = False,
+  skip_ts_build: Optional[bool] = False,
+  skip_data_upload: Optional[bool] = False,
+  create_space: Optional[bool] = False,
+) -> None:
   """Generate the huggingface space app."""
   hf_space = hf_space or env('HF_STAGING_DEMO_REPO')
   if not hf_space:
@@ -119,7 +128,9 @@ def deploy_staging(hf_space: Optional[str] = None,
       make_datasets_public=False,
       skip_data_upload=skip_data_upload,
       hf_space_storage=None,
-      create_space=create_space))
+      create_space=create_space,
+    )
+  )
 
   # Unconditionally remove dist. dist is unconditionally uploaded so it is empty when using
   # the public package.
@@ -143,7 +154,9 @@ def deploy_staging(hf_space: Optional[str] = None,
     operations.append(
       CommitOperationAdd(
         path_in_repo=os.path.join(PY_DIST_DIR, upload_file),
-        path_or_fileobj=os.path.join(PY_DIST_DIR, upload_file)))
+        path_or_fileobj=os.path.join(PY_DIST_DIR, upload_file),
+      )
+    )
 
   # Atomically commit all the operations so we don't kick the server multiple times.
   hf_api.create_commit(

web/blueprint/src/lib/components/NavigationGroup.svelte

@@ -27,6 +27,8 @@
   export let isFetching: boolean;
   export let tagGroups: NavigationTagGroup[];
   export let expanded = true;
+
+  $: hasTags = tagGroups.some(({tag}) => tag != '');
 </script>
 
 <div class="my-1 w-full px-1">
@@ -51,13 +53,17 @@
       {:else}
         <div class="mt-1">
           {#each tagGroups as { tag, groups }}
-            {#if tag != ''}
+            {#if hasTags}
               <div
                 class="flex flex-row justify-between pl-3
             text-sm opacity-80"
               >
                 <div class="py-1 text-xs">
-                  <Tag type="purple" size="sm">{tag}</Tag>
+                  {#if tag != ''}
+                    <Tag type="purple" size="sm">{tag}</Tag>
+                  {:else}
+                    <Tag type="cool-gray" size="sm">no tag</Tag>
+                  {/if}
                 </div>
               </div>
             {/if}

web/blueprint/src/lib/components/commands/selectors/CommandSelectList.svelte

@@ -22,7 +22,7 @@
       <button
         data-active={item === _item.value}
         on:click={() => dispatch('select', _item.value)}
-        class="mb-1 flex items-center justify-between"
+        class="my-1 flex items-center justify-between"
       >
         {_item.title}
         {#if _item.tag} <Tag size="sm" type={_item.tag.type}>{_item.tag.value}</Tag> {/if}

web/blueprint/src/lib/components/datasetView/DatasetSettingsFields.svelte

@@ -76,6 +76,7 @@
   $: {
     if (selectedMediaFields != null) {
       settings = {
+        ...settings,
         ui: {
           media_paths: selectedMediaFields.map(f => f.path),
           markdown_paths: markdownMediaFields?.map(f => f.path)