Actualiser automatiquement la base Hugging Face

.gitattributes

@@ -42,4 +42,3 @@ application_neo4j/static/notice/notice_html/media/image4.png filter=lfs diff=lfs
 application_neo4j/static/notice/notice.docx filter=lfs diff=lfs merge=lfs -text
 application_neo4j/static/notice/notice.pdf filter=lfs diff=lfs merge=lfs -text
 opengds/*.jar filter=lfs diff=lfs merge=lfs -text
-application_neo4j/static/notice/notice_en.pdf filter=lfs diff=lfs merge=lfs -text

application_neo4j/app.py

@@ -24,7 +24,6 @@ app.secret_key = os.urandom(24)
 # --- Connexion à Neo4j et GDS ---
 NEO4J_URI = "bolt://localhost:7687"
 GDS_GRAPH_NAME = "genealogie_gds"
-GDS_RELATIONSHIP_TYPES = ("IS_IN", "POSTED", "USED_IN")
 SEARCH_JOB_TTL_SECONDS = 60 * 60
 RESULT_BATCH_SIZE = 25
 DEFAULT_SEARCH_MAX_WORKERS = 2
@@ -149,38 +148,6 @@ def ensure_graph_projected(
         if g_reverse_exists:
             gds.graph.get(reverse_graph_name).drop()
 
-        relationship_types_result = gds.run_cypher(
-            """
-            CALL db.relationshipTypes()
-            YIELD relationshipType
-            RETURN relationshipType
-            """
-        )
-        available_relationship_types = set(
-            relationship_types_result["relationshipType"].tolist()
-        )
-        projected_relationship_types = [
-            relationship_type
-            for relationship_type in GDS_RELATIONSHIP_TYPES
-            if relationship_type in available_relationship_types
-        ]
-        if not projected_relationship_types:
-            raise RuntimeError(
-                "Aucun type de relation compatible avec la recherche GDS "
-                "n'est présent dans Neo4j."
-            )
-        reverse_relationship_projection = {
-            relationship_type: {
-                "type": relationship_type,
-                "orientation": "REVERSE",
-            }
-            for relationship_type in projected_relationship_types
-        }
-        print(
-            "Types de relations projetés dans GDS : "
-            + ", ".join(projected_relationship_types)
-        )
-
         # Native projections accept a jobId, allowing gds.listProgress to
         # expose real progress while the first search prepares the graph.
         natural_projection_job_id = (
@@ -196,7 +163,7 @@ def ensure_graph_projected(
             CALL gds.graph.project(
                 $graph_name,
                 '*',
-                $relationship_projection,
+                ['IS_IN', 'POSTED', 'USED_IN'],
                 {jobId: $job_id}
             )
             YIELD graphName, nodeCount, relationshipCount
@@ -205,7 +172,6 @@ def ensure_graph_projected(
             {
                 "graph_name": natural_graph_name,
                 "job_id": natural_projection_job_id,
-                "relationship_projection": projected_relationship_types,
             },
         )
         print(f"Graphe '{natural_graph_name}' projeté.")
@@ -223,7 +189,11 @@ def ensure_graph_projected(
             CALL gds.graph.project(
                 $graph_name,
                 '*',
-                $relationship_projection,
+                {
+                    IS_IN: {type: 'IS_IN', orientation: 'REVERSE'},
+                    POSTED: {type: 'POSTED', orientation: 'REVERSE'},
+                    USED_IN: {type: 'USED_IN', orientation: 'REVERSE'}
+                },
                 {jobId: $job_id}
             )
             YIELD graphName, nodeCount, relationshipCount
@@ -232,7 +202,6 @@ def ensure_graph_projected(
             {
                 "graph_name": reverse_graph_name,
                 "job_id": reverse_projection_job_id,
-                "relationship_projection": reverse_relationship_projection,
             },
         )
         print(f"Graphe '{reverse_graph_name}' projeté.")
@@ -287,11 +256,6 @@ def inject_i18n():
         "search.progress_stage_completed", "search.progress_stage_failed",
         "search.progress_stage_cancelled",
         "search.progress_elapsed", "search.progress_items",
-        "search.progress_queue_title",
-        "search.progress_queue_position", "search.progress_queue_ahead_one",
-        "search.progress_queue_ahead_many", "search.progress_queue_next",
-        "search.progress_queue_running_one", "search.progress_queue_running_many",
-        "search.progress_queue_note",
         "search.progress_server_percent",
         "search.progress_remaining_step", "search.progress_server_note",
         "search.progress_connection_error",
@@ -326,16 +290,13 @@ def autocomplete():
     if node_filter and node_filter in ["Model", "Dataset"]: # Mesure de sécurité
         label_cypher = f":{node_filter}"
 
-    # Récupère les noms commençant par le préfixe fourni. Les suggestions les
-    # plus téléchargées sont proposées en premier ; le nom garantit un ordre
-    # stable lorsque plusieurs éléments ont le même nombre de téléchargements.
+    # Récupère les noms commençant par le préfixe fourni
     cypher = f"""
         MATCH (n{label_cypher})
         WHERE toLower(n.name) STARTS WITH toLower($prefix)
         AND n.name IS NOT NULL
-        RETURN n.name AS name, labels(n)[0] AS label,
-               coalesce(n.downloads, 0) AS downloads
-        ORDER BY downloads DESC, toLower(n.name) ASC
+        RETURN n.name AS name, labels(n)[0] as label
+        ORDER BY size(n.name) ASC
         LIMIT 10
     """
     try:
@@ -408,8 +369,6 @@ def make_search_result(name, depth, is_unlimited, filters, expert):
     return {
         "template": "expert.html" if expert else "search.html",
         "message": None,
-        "message_key": None,
-        "message_kwargs": {},
         "search": {
             "name": name,
             "depth": depth,
@@ -421,13 +380,6 @@ def make_search_result(name, depth, is_unlimited, filters, expert):
     }
 
 
-def set_search_result_message(result, key, lang, **kwargs):
-    """Store a translatable message while keeping its current rendering."""
-    result["message_key"] = key
-    result["message_kwargs"] = kwargs
-    result["message"] = t(key, lang, **kwargs)
-
-
 def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang):
     """Run the existing search pipeline while publishing its real server stage."""
     result = make_search_result(name, depth, is_unlimited, filters, expert)
@@ -458,7 +410,6 @@ def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang)
             name,
             None if is_unlimited else depth,
             expert,
-            source_labels=filters,
             job_id=job_id,
             progress_callback=report_gds_stage,
             neo4j_driver=driver,
@@ -466,8 +417,7 @@ def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang)
         )
 
         if not gds_result:
-            set_search_result_message(
-                result,
+            result["message"] = t(
                 "error.node_not_found_expert" if expert else "error.model_not_found",
                 lang,
                 name=name,
@@ -489,9 +439,7 @@ def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang)
 
             if expert:
                 if not graph_data["nodes"] and not graph_data["edges"]:
-                    set_search_result_message(
-                        result, "error.no_neighbors", lang, name=name
-                    )
+                    result["message"] = t("error.no_neighbors", lang, name=name)
             elif gds_result["source_label"] == "Model":
                 raise_if_search_cancelled(job_id)
                 set_search_stage(job_id, "building_highlights")
@@ -501,9 +449,7 @@ def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang)
             elif gds_result["source_label"] == "Dataset":
                 result["template"] = "search_dataset.html"
             elif not graph_data["nodes"] and not graph_data["edges"]:
-                set_search_result_message(
-                    result, "error.no_neighbors", lang, name=name
-                )
+                result["message"] = t("error.no_neighbors", lang, name=name)
 
         raise_if_search_cancelled(job_id)
         update_search_job(
@@ -544,15 +490,11 @@ def execute_search_job(job_id, name, depth, is_unlimited, filters, expert, lang)
             )
             return
         if "Failed to find a node" in str(error):
-            message_key = "error.node_not_found"
-            message_kwargs = {"name": name}
+            message = t("error.node_not_found", lang, name=name)
         else:
             print(f"Background GDS search error ({job_id}): {error}")
-            message_key = "error.gds"
-            message_kwargs = {"error": str(error)}
-        set_search_result_message(
-            result, message_key, lang, **message_kwargs
-        )
+            message = t("error.gds", lang, error=str(error))
+        result["message"] = message
         update_search_job(
             job_id,
             status="failed",
@@ -737,25 +679,6 @@ def search_job_status(job_id):
     with search_jobs_lock:
         stored_job = search_jobs.get(job_id)
         job = dict(stored_job) if stored_job else None
-        queued_jobs = sorted(
-            (
-                (queued_job.get("created_at", 0), queued_job_id)
-                for queued_job_id, queued_job in search_jobs.items()
-                if queued_job.get("status") == "queued"
-            )
-        )
-        queued_job_ids = [
-            queued_job_id for _, queued_job_id in queued_jobs
-        ]
-        queue_position = (
-            queued_job_ids.index(job_id) + 1
-            if job_id in queued_job_ids
-            else None
-        )
-        running_jobs = sum(
-            queued_job.get("status") == "running"
-            for queued_job in search_jobs.values()
-        )
     if not job:
         return jsonify({"error": "Search job not found"}), 404
 
@@ -767,9 +690,6 @@ def search_job_status(job_id):
         "elapsed_seconds": max(0, math.floor(now - started_at)) if started_at else 0,
         "progress_percent": None,
         "remaining_seconds": None,
-        "queue_position": queue_position,
-        "queued_jobs": len(queued_job_ids),
-        "running_jobs": running_jobs,
         "completed_items": job.get("completed_items"),
         "total_items": job.get("total_items"),
         "result_url": (
@@ -808,16 +728,10 @@ def search_job_result(job_id):
     if job["status"] not in ("completed", "failed") or not job.get("result"):
         return redirect(url_for("findnode", lang=job["lang"]))
 
+    session["lang"] = job["lang"]
     result = job["result"]
-    message = result["message"]
-    if result.get("message_key"):
-        message = t(
-            result["message_key"],
-            session.get("lang", "fr"),
-            **result.get("message_kwargs", {}),
-        )
     template_args = {
-        "message": message,
+        "message": result["message"],
         "search": result["search"],
         "graph_data": result["graph_data"],
     }
@@ -882,7 +796,6 @@ def findnode():
                 name,
                 depth,
                 False,
-                source_labels=current_filters,
             )
             if not gds_result :
                 message = t("error.model_not_found", session.get("lang", "fr"), name=name)
@@ -964,7 +877,6 @@ def findnode_expert():
                 name,
                 depth,
                 True,
-                source_labels=current_filters,
             )
             if not gds_result :
                 message = t("error.node_not_found_expert", session.get("lang", "fr"), name=name)

application_neo4j/app_algorithms.py

@@ -12,7 +12,6 @@ def run_gds_bfs(
     source_name: str,
     max_depth: int = None,
     expert=False,
-    source_labels=None,
     job_id: str = None,
     progress_callback=None,
     neo4j_driver=None,
@@ -29,55 +28,26 @@ def run_gds_bfs(
         L'identifiant et le label de la source, ainsi que les résultats GDS
         des parcours descendant et ascendant.
     """
-    requested_source_labels = [
-        label
-        for label in (source_labels or [])
-        if label in ("Model", "Dataset", "Author")
-    ]
     try:
         source_id_result = gds.run_cypher(
             """
             MATCH (n {name: $source_name})
-            WHERE size($source_labels) = 0
-               OR any(label IN labels(n) WHERE label IN $source_labels)
             RETURN id(n) AS id, labels(n) AS label
+            LIMIT 1
             """,
-            {
-                "source_name": source_name,
-                "source_labels": requested_source_labels,
-            },
+            {"source_name": source_name},
         )
 
         if source_id_result.empty:
             print(f"Le modèle ou dataset avec le nom '{source_name}' n'a pas été trouvé.")
             return None
 
-        label_preferences = requested_source_labels or [
-            "Model",
-            "Dataset",
-            "Author",
-        ]
-        selected_source = None
-        source_label = None
-        for preferred_label in label_preferences:
-            matching_sources = source_id_result[
-                source_id_result["label"].apply(
-                    lambda node_labels: preferred_label in node_labels
-                )
-            ]
-            if not matching_sources.empty:
-                selected_source = matching_sources.iloc[0]
-                source_label = preferred_label
-                break
-        if selected_source is None:
-            selected_source = source_id_result.iloc[0]
-            source_label = selected_source["label"][0]
-
+        source_label = source_id_result["label"].iloc[0][0]
         if source_label == "Author" and not expert:
             print(f"Le modèle ou dataset avec le nom '{source_name}' n'a pas été trouvé.")
             return None
 
-        source_node_id = int(selected_source["id"])
+        source_node_id = int(source_id_result["id"].iloc[0])
     except Exception as e:
         print(f"Erreur lors de la recherche de l'ID du nœud source pour '{source_name}': {e}")
         return None
@@ -186,50 +156,50 @@ def get_genealogy_highlights(gds: "GraphDataScience", model_name: str, num_highl
     # Les classes CSS sont directement des classes Bootstrap 5 pour simplifier le rendu dans le template HTML.
     badges_info = {
     'desc_cited_1': {
-        'text_key': 'badge.desc_cited_1.text',
+        'text': t('badge.desc_cited_1.text', lang), 
         'class': 'bg-success',
-        'title_key': 'badge.desc_cited_1.title'
+        'title': t('badge.desc_cited_1.title', lang)
     },
     'desc_cited_2': {
-        'text_key': 'badge.desc_cited_2.text',
+        'text': t('badge.desc_cited_2.text', lang), 
         'class': 'bg-success bg-opacity-75',
-        'title_key': 'badge.desc_cited_2.title'
+        'title': t('badge.desc_cited_2.title', lang)
     },
     'desc_downloaded_1': {
-        'text_key': 'badge.desc_downloaded_1.text',
+        'text': t('badge.desc_downloaded_1.text', lang), 
         'class': 'beta',
-        'title_key': 'badge.desc_downloaded_1.title'
+        'title': t('badge.desc_downloaded_1.title', lang)
     },
     'desc_downloaded_2': {
-        'text_key': 'badge.desc_downloaded_2.text',
+        'text': t('badge.desc_downloaded_2.text', lang), 
         'class': 'alpha',
-        'title_key': 'badge.desc_downloaded_2.title'
+        'title': t('badge.desc_downloaded_2.title', lang)
     },
     
     'asc_foundation': {
-        'text_key': 'badge.asc_foundation.text',
+        'text': t('badge.asc_foundation.text', lang), 
         'class': 'bg-warning text-dark',
-        'title_key': 'badge.asc_foundation.title'
+        'title': t('badge.asc_foundation.title', lang)
     },
     'asc_cited_1': {
-        'text_key': 'badge.asc_cited_1.text',
+        'text': t('badge.asc_cited_1.text', lang), 
         'class': 'bg-success',
-        'title_key': 'badge.asc_cited_1.title'
+        'title': t('badge.asc_cited_1.title', lang)
     },
     'asc_cited_2': {
-        'text_key': 'badge.asc_cited_2.text',
+        'text': t('badge.asc_cited_2.text', lang), 
         'class': 'bg-success bg-opacity-75',
-        'title_key': 'badge.asc_cited_2.title'
+        'title': t('badge.asc_cited_2.title', lang)
     },
     'asc_downloaded_1': {
-        'text_key': 'badge.asc_downloaded_1.text',
+        'text': t('badge.asc_downloaded_1.text', lang), 
         'class': 'beta',
-        'title_key': 'badge.asc_downloaded_1.title'
+        'title': t('badge.asc_downloaded_1.title', lang)
     },
     'asc_downloaded_2': {
-        'text_key': 'badge.asc_downloaded_2.text',
+        'text': t('badge.asc_downloaded_2.text', lang), 
         'class': 'alpha',
-        'title_key': 'badge.asc_downloaded_2.title'
+        'title': t('badge.asc_downloaded_2.title', lang)
     },
     }
     

application_neo4j/static/css/style.css

@@ -2,79 +2,13 @@
  *  STYLE GLOBAL & GRAPHIQUE SIGMA
  * =================================================================== */
 
-/* Explicit sizing and minimum widths avoid engine-specific flexbox
- * expansion differences (notably between Blink and Gecko). */
-html {
-  -webkit-text-size-adjust: 100%;
-  text-size-adjust: 100%;
-}
-
-body {
-  min-width: 320px;
-}
-
-.legacy-browser-warning {
-  display: none;
-  margin: 0;
-  padding: 0.75rem 1rem;
-  color: #664d03;
-  background: #fff3cd;
-  border-bottom: 1px solid #ffecb5;
-  font-weight: 600;
-  text-align: center;
-}
-
-img,
-svg {
-  max-width: 100%;
-  height: auto;
-}
-
-.row > *,
-.card,
-.card-body,
-.flex-grow-1 {
-  min-width: 0;
-}
-
-.navbar .container {
-  column-gap: 1rem;
-  row-gap: 0.75rem;
-}
-
-.navbar-brand {
-  min-width: 0;
-  max-width: 100%;
-  white-space: normal;
-}
-
-.navbar-brand small,
-#suggestions-list li {
-  overflow-wrap: anywhere;
-  word-break: break-word;
-}
-
-.navbar-actions {
-  min-width: 0;
-  margin-left: auto;
-}
-
  #sigma-container {
   width: 100%;
   height: 600px; /* Un peu plus de hauteur pour un meilleur confort */
-  position: relative;
-  overflow: hidden;
-  border: 1px solid #dee2e6;
-  border: 1px solid var(--bs-border-color, #dee2e6); /* Variable Bootstrap */
+  border: 1px solid var(--bs-border-color); /* Variable Bootstrap */
   margin-top: 1rem;
-  border-radius: 0.375rem;
-  border-radius: var(--bs-border-radius, 0.375rem); /* Variable Bootstrap */
-  background-color: #f8f9fa;
-  background-color: var(--bs-light-bg-subtle, #f8f9fa); /* Fond légèrement teinté */
-}
-
-#floating-legend {
-  max-width: calc(100% - 1rem);
+  border-radius: var(--bs-border-radius); /* Variable Bootstrap */
+  background-color: var(--bs-light-bg-subtle); /* Fond légèrement teinté */
 }
 
 /* ===================================================================
@@ -98,62 +32,35 @@ svg {
 /* ===================================================================
  *  STYLE DES TABLES (DATATABLES)
  * =================================================================== */
-/* A single, deterministic wrapping policy replaces the previous contradictory
- * normal/nowrap declarations. */
+/* Permet au texte de revenir à la ligne naturellement */
 #descendance-table th, #descendance-table td,
-#ascendance-table th, #ascendance-table td,
-#train-table th, #train-table td,
-#graph-models-table th, #graph-models-table td {
-    white-space: nowrap;
-    vertical-align: middle;
+#ascendance-table th, #ascendance-table td {
+    white-space: normal; /* <-- La correction clé ! */
+    word-wrap: break-word; /* Force le retour à la ligne des mots longs */
+    vertical-align: middle; /* Garde le centrage vertical */
 }
 
-#descendance-table th,
-#ascendance-table th,
-#train-table th {
-    min-width: 120px;
+/* On peut donner une largeur minimale aux colonnes pour garder une bonne structure */
+#descendance-table th, #ascendance-table th {
+    min-width: 120px; /* Ajustez cette valeur selon vos besoins */
 }
 
-#descendance-table th:first-child, #descendance-table td:first-child,
-#ascendance-table th:first-child, #ascendance-table td:first-child,
-#train-table th:first-child, #train-table td:first-child,
-#graph-models-table th:first-child, #graph-models-table td:first-child {
+/* Cas spécifique pour les colonnes potentiellement longues comme le nom du modèle */
+#descendance-table th:first-child, #ascendance-table th:first-child {
     min-width: 200px;
-    white-space: normal;
-    overflow-wrap: anywhere;
-    word-break: break-word;
 }
 
-.table-responsive,
-div.dataTables_wrapper,
-div.dataTables_wrapper div.dataTables_scroll {
-  width: 100%;
-  max-width: 100%;
-}
-
-.table-responsive,
-div.dataTables_wrapper div.dataTables_scrollBody {
-  overflow-x: auto !important;
-  -webkit-overflow-scrolling: touch;
-}
-
-div.dataTables_wrapper div.dataTables_scrollHeadInner {
-  min-width: 100%;
-}
-
-div.dataTables_wrapper table.dataTable {
-  margin-left: 0 !important;
-  margin-right: 0 !important;
-  table-layout: auto;
+/* Empêche le texte de se couper dans les cellules, force le défilement horizontal */
+#descendance-table th, #descendance-table td,
+#ascendance-table th, #ascendance-table td {
+    white-space: nowrap;
+    vertical-align: middle; /* Centrage vertical pour un look plus propre */
 }
 
 /* Personnalisation de l'input de recherche de DataTables pour qu'il ressemble à un champ Bootstrap */
 div.dataTables_wrapper div.dataTables_filter input {
-  max-width: 100%;
-  border: 1px solid #dee2e6;
-  border: 1px solid var(--bs-border-color, #dee2e6);
-  border-radius: 0.375rem;
-  border-radius: var(--bs-border-radius, 0.375rem);
+  border: 1px solid var(--bs-border-color);
+  border-radius: var(--bs-border-radius);
   padding: 0.375rem 0.75rem;
   margin-left: 0.5em;
 }
@@ -193,12 +100,8 @@ div.dataTables_wrapper .table thead th.sorting_desc::after {
   /* Ce style est pour si vous utilisez une <ul> simple.
      Si vous avez bien une <ul class="list-group">, vous pouvez supprimer ce bloc. */
   box-shadow: 0 0.5rem 1rem rgba(0, 0, 0, 0.15);
-  max-width: 100%;
-  overflow-x: hidden;
-  border-radius: 0.375rem;
-  border-radius: var(--bs-border-radius, 0.375rem);
-  background-color: #f8f9fa;
-  background-color: var(--bs-light, #f8f9fa);
+  border-radius: var(--bs-border-radius);
+  background-color: var(--bs-light);
 }
 #suggestions-list li:hover {
   background-color:  #94bfeb; ;
@@ -295,22 +198,11 @@ div.dataTables_wrapper .table thead th.sorting_desc::after {
 }
 
 /* Change la couleur de fond des cases à cocher lorsqu'elles sont cochées */
-.btn-primary {
+.btn {
   background-color: #6a11cb; 
   border-color:#6a11cb;    
 }
 
-.btn-outline-primary {
-  color: #6a11cb;
-  border-color: #6a11cb;
-}
-
-.btn-outline-primary:hover {
-  color: #fff;
-  background-color: #6a11cb;
-  border-color: #6a11cb;
-}
-
 .stretched-link{
 color: #6a11cb; 
 }
@@ -376,31 +268,4 @@ main {
 .btn-modern:hover {
   background: linear-gradient(135deg, #5a0fbf, #1d63e1);
   transform: translateY(-3px);
-}
-
-@media (max-width: 575.98px) {
-  .navbar-brand,
-  .navbar-actions {
-    flex: 1 1 100%;
-    margin-right: 0;
-  }
-
-  .navbar-actions {
-    justify-content: flex-start !important;
-  }
-
-  #sigma-container {
-    height: 70vh;
-    min-height: 420px;
-  }
-
-  div.dataTables_wrapper div.dataTables_filter label {
-    display: block;
-    width: 100%;
-  }
-
-  div.dataTables_wrapper div.dataTables_filter input {
-    width: 100%;
-    margin: 0.375rem 0 0;
-  }
-}
+}

application_neo4j/static/js/browser_compatibility.js

@@ -1,14 +0,0 @@
-(function () {
-  "use strict";
-
-  // `document.documentMode` is exposed by Internet Explorer only. The rest of
-  // the application deliberately targets maintained browsers supported by
-  // Bootstrap 5.
-  if (document.documentMode) {
-    var warning = document.getElementById("legacy-browser-warning");
-    if (warning) {
-      warning.hidden = false;
-      warning.style.display = "block";
-    }
-  }
-}());

application_neo4j/static/js/script.js

@@ -193,15 +193,12 @@ document.addEventListener("DOMContentLoaded", () => {
             const common_dt_options = {
                 "language": { "url": datatablesLangUrl },
                 "pageLength": 10,
-                "responsive": true,
-                "scrollX": true,
-                // Afficher en premier les modèles les plus téléchargés.
-                "order": [[2, "desc"]],
+                "responsive": true,"scrollX": true , "scrollY":true,
                 "columnDefs": [
                 {
                     // Appliquer notre plugin 'numeric-string' aux colonnes cibles
                     "type": "numeric-string", 
-                    "targets": [2, 4, 8, 9, 10, 11] // Downloads, Likes, Distance, Ascendants, Descendants, Citations
+                    "targets": [2, 4, 6, 7, 8, 9] // Indices des colonnes: Downloads, Likes, Distance, etc.
                 }
             ]
                 // "searching" est activé par défaut, on peut le customiser

application_neo4j/static/js/script_dataset.js

@@ -209,8 +209,7 @@ document.addEventListener("DOMContentLoaded", () => {
             const common_dt_options = {
                 "language": { "url": datatablesLangUrl },
                 "pageLength": 10,
-                "responsive": true,
-                "scrollX": true,
+                "responsive": true,"scrollX": true , "scrollY":true,
                 "columnDefs": [
                 {
                     // Appliquer notre plugin 'numeric-string' aux colonnes cibles

application_neo4j/static/js/search_progress.js

@@ -95,7 +95,6 @@
   }
 
   function renderStatus(progress, job) {
-    const title = progress.querySelector(".search-progress-title");
     const stage = progress.querySelector(".search-progress-stage");
     const status = progress.querySelector(".search-progress-status");
     const note = progress.querySelector(".search-progress-note");
@@ -103,84 +102,6 @@
 
     stage.textContent = stageLabel(job.stage);
 
-    if (job.status === "queued" || job.stage === "queued") {
-      title.textContent = translate(
-        "search.progress_queue_title",
-        "Recherche en file d’attente"
-      );
-      progress.classList.remove("alert-info", "alert-danger", "alert-secondary");
-      progress.classList.add("alert-warning");
-
-      const queueDetails = [];
-      if (job.queue_position && job.queued_jobs) {
-        queueDetails.push(
-          format(
-            translate(
-              "search.progress_queue_position",
-              "Position dans la file : {position}/{total}"
-            ),
-            {
-              position: job.queue_position,
-              total: job.queued_jobs,
-            }
-          )
-        );
-        const jobsAhead = job.queue_position - 1;
-        if (jobsAhead === 0) {
-          queueDetails.push(
-            translate(
-              "search.progress_queue_next",
-              "Votre recherche est la prochaine."
-            )
-          );
-        } else {
-          queueDetails.push(
-            format(
-              translate(
-                jobsAhead === 1
-                  ? "search.progress_queue_ahead_one"
-                  : "search.progress_queue_ahead_many",
-                jobsAhead === 1
-                  ? "1 recherche devant la vôtre"
-                  : "{count} recherches devant la vôtre"
-              ),
-              { count: jobsAhead }
-            )
-          );
-        }
-      }
-      if (job.running_jobs) {
-        queueDetails.push(
-          format(
-            translate(
-              job.running_jobs === 1
-                ? "search.progress_queue_running_one"
-                : "search.progress_queue_running_many",
-              job.running_jobs === 1
-                ? "1 recherche en cours"
-                : "{count} recherches en cours"
-            ),
-            { count: job.running_jobs }
-          )
-        );
-      }
-      status.textContent = queueDetails.join(" · ");
-      note.textContent = translate(
-        "search.progress_queue_note",
-        "La recherche démarrera automatiquement dès qu’une capacité sera disponible."
-      );
-      bar.style.width = "0%";
-      bar.removeAttribute("aria-valuenow");
-      return;
-    }
-
-    title.textContent = translate(
-      "search.progress_title",
-      "Recherche en cours…"
-    );
-    progress.classList.remove("alert-warning", "alert-danger", "alert-secondary");
-    progress.classList.add("alert-info");
-
     const details = [
       format(
         translate("search.progress_elapsed", "Temps écoulé : {seconds} s"),
@@ -287,9 +208,6 @@
       elapsed_seconds: 0,
       progress_percent: null,
       remaining_seconds: null,
-      queue_position: null,
-      queued_jobs: null,
-      running_jobs: null,
       completed_items: null,
       total_items: null,
     });

application_neo4j/static/js/utils.js

@@ -231,9 +231,6 @@ function formatDateFr(dateString) {
 
 // Fonction pour parser la valeur : enlève les espaces et convertit en nombre si possible
 function parseNumericValue(value) {
-    if (typeof value === 'number' && Number.isFinite(value)) {
-        return { isNumber: true, value };
-    }
     if (typeof value === 'string') {
         // Enlève les espaces (pour les nombres comme "12 345") et les virgules
         const cleanedValue = value.replace(/[\s,]/g, '');
@@ -247,7 +244,7 @@ function parseNumericValue(value) {
 }
 
 // Définition du tri ascendant
-jQuery.fn.dataTable.ext.type.order['numeric-string-asc'] = function (a, b) {
+jQuery.fn.dataTable.ext.order['numeric-string-asc'] = function (a, b) {
     const valA = parseNumericValue(a);
     const valB = parseNumericValue(b);
 
@@ -263,19 +260,9 @@ jQuery.fn.dataTable.ext.type.order['numeric-string-asc'] = function (a, b) {
     }
 };
 
-// Tri descendant, en conservant les valeurs inconnues après les nombres.
-jQuery.fn.dataTable.ext.type.order['numeric-string-desc'] = function (a, b) {
-    const valA = parseNumericValue(a);
-    const valB = parseNumericValue(b);
-
-    if (valA.isNumber && valB.isNumber) {
-        return valB.value - valA.value;
-    } else if (valA.isNumber && !valB.isNumber) {
-        return -1;
-    } else if (!valA.isNumber && valB.isNumber) {
-        return 1;
-    }
-    return String(valB.value).localeCompare(String(valA.value));
+// Le tri descendant est simplement l'inverse de l'ascendant
+jQuery.fn.dataTable.ext.order['numeric-string-desc'] = function (a, b) {
+    return jQuery.fn.dataTable.ext.order['numeric-string-asc'](a, b) * -1;
 };
 
 /**

application_neo4j/static/notice/generate_notice_en.py

@@ -1,503 +0,0 @@
-"""Generate the English version of the CNIL GenMod information notice.
-
-This is a maintainer utility, not a runtime dependency of the Space. Run it
-from the repository root after installing ``pypdf`` and ``reportlab``:
-
-    python application_neo4j/static/notice/generate_notice_en.py \
-        --source application_neo4j/static/notice/notice.pdf
-"""
-
-from __future__ import annotations
-
-import argparse
-import math
-import tempfile
-from pathlib import Path
-
-from PIL import Image as PillowImage
-from pypdf import PdfReader
-from reportlab.lib import colors
-from reportlab.lib.enums import TA_CENTER
-from reportlab.lib.pagesizes import A4
-from reportlab.lib.styles import ParagraphStyle, getSampleStyleSheet
-from reportlab.lib.units import mm
-from reportlab.pdfbase import pdfmetrics
-from reportlab.pdfbase.ttfonts import TTFont
-from reportlab.platypus import (
-    Flowable,
-    Image,
-    PageBreak,
-    Paragraph,
-    SimpleDocTemplate,
-    Spacer,
-    Table,
-    TableStyle,
-)
-
-
-CNIL_BLUE = colors.HexColor("#0046A8")
-LIGHT_BLUE = colors.HexColor("#EAF2FB")
-RED = colors.HexColor("#E52A22")
-GOLD = colors.HexColor("#F7B500")
-
-
-class NeuralNetworkDiagram(Flowable):
-    """Small English recreation of the neural-network diagram."""
-
-    def __init__(self, width=170 * mm, height=54 * mm):
-        super().__init__()
-        self.width = width
-        self.height = height
-
-    def draw(self):
-        canvas = self.canv
-        canvas.saveState()
-        scale_x = self.width / 480
-        scale_y = self.height / 155
-        canvas.scale(scale_x, scale_y)
-
-        inputs = [(45, 115), (45, 78), (45, 41)]
-        hidden_1 = [(190, 130), (190, 97), (190, 64), (190, 31)]
-        hidden_2 = [(330, 130), (330, 97), (330, 64), (330, 31)]
-        outputs = [(445, 113), (445, 78), (445, 43)]
-
-        canvas.setStrokeColor(colors.HexColor("#222222"))
-        canvas.setLineWidth(0.6)
-        for x1, y1 in inputs:
-            for x2, y2 in hidden_1:
-                canvas.line(x1 + 32, y1, x2 - 10, y2)
-        for x1, y1 in hidden_2:
-            for x2, y2 in outputs:
-                canvas.line(x1 + 10, y1, x2 - 32, y2)
-
-        canvas.setFont("DejaVuSans", 7.5)
-        for index, (x, y) in enumerate(inputs, 1):
-            canvas.setFillColor(colors.HexColor("#DDEFD6"))
-            canvas.roundRect(x - 32, y - 10, 64, 20, 4, fill=1, stroke=1)
-            canvas.setFillColor(colors.black)
-            canvas.drawCentredString(x, y - 3, f"Input {index}")
-        for layer in (hidden_1, hidden_2):
-            for x, y in layer:
-                canvas.setFillColor(colors.HexColor("#FFF1C9"))
-                canvas.circle(x, y, 10, fill=1, stroke=1)
-        for index, (x, y) in enumerate(outputs, 1):
-            canvas.setFillColor(colors.HexColor("#DCE9FF"))
-            canvas.roundRect(x - 32, y - 10, 64, 20, 4, fill=1, stroke=1)
-            canvas.setFillColor(colors.black)
-            canvas.drawCentredString(x, y - 3, f"Output {index}")
-
-        canvas.setFillColor(colors.white)
-        path = canvas.beginPath()
-        path.moveTo(255, 80)
-        path.lineTo(275, 98)
-        path.lineTo(295, 80)
-        path.lineTo(275, 62)
-        path.close()
-        canvas.drawPath(path, fill=1, stroke=1)
-        canvas.setFillColor(colors.black)
-        canvas.setFont("DejaVuSans", 10)
-        canvas.drawCentredString(275, 77, "f")
-        canvas.setFont("DejaVuSans", 7.5)
-        canvas.drawCentredString(190, 8, "Layer 1")
-        canvas.drawCentredString(330, 8, "Layer 2")
-        canvas.drawCentredString(275, 118, "Activation function")
-        canvas.drawCentredString(405, 145, "Edge parameters")
-        canvas.restoreState()
-
-
-class HuggingFaceDiagram(Flowable):
-    """English recreation of the platform example diagram."""
-
-    def __init__(self, width=170 * mm, height=55 * mm):
-        super().__init__()
-        self.width = width
-        self.height = height
-
-    def draw_arrow(self, canvas, x1, y1, x2, y2, color):
-        canvas.setStrokeColor(color)
-        canvas.setFillColor(color)
-        canvas.setLineWidth(1.8)
-        canvas.line(x1, y1, x2, y2)
-        angle = math.atan2(y2 - y1, x2 - x1)
-        for offset in (-0.5, 0.5):
-            canvas.line(
-                x2,
-                y2,
-                x2 - 7 * math.cos(angle + offset),
-                y2 - 7 * math.sin(angle + offset),
-            )
-
-    def box(self, canvas, x, y, width, height, text, color):
-        canvas.setStrokeColor(color)
-        canvas.setFillColor(colors.white)
-        canvas.rect(x, y, width, height, fill=1, stroke=1)
-        canvas.setFillColor(color)
-        canvas.setFont("DejaVuSans-Bold", 8)
-        canvas.drawCentredString(x + width / 2, y + height / 2 - 3, text)
-
-    def draw(self):
-        canvas = self.canv
-        canvas.saveState()
-        scale_x = self.width / 500
-        scale_y = self.height / 170
-        canvas.scale(scale_x, scale_y)
-        self.box(canvas, 5, 118, 70, 24, "USER A", CNIL_BLUE)
-        self.box(canvas, 105, 105, 105, 48, "", RED)
-        canvas.setFillColor(RED)
-        canvas.setFont("DejaVuSans-Bold", 8)
-        canvas.drawCentredString(157.5, 130, "MACHINE-LEARNING")
-        canvas.drawCentredString(157.5, 116, "MODEL")
-        self.box(canvas, 290, 105, 85, 48, "DATASET", RED)
-        self.box(canvas, 420, 118, 75, 24, "USER B", CNIL_BLUE)
-        self.box(canvas, 205, 15, 90, 26, "USER C", CNIL_BLUE)
-        self.box(canvas, 207, 66, 86, 25, "NEW MODEL", RED)
-
-        canvas.setFillColor(GOLD)
-        canvas.circle(250, 130, 21, fill=1, stroke=0)
-        canvas.setFillColor(colors.HexColor("#795500"))
-        canvas.setFont("DejaVuSans-Bold", 8)
-        canvas.drawCentredString(250, 127, "HF")
-        canvas.setFont("DejaVuSans-Bold", 7)
-        canvas.drawCentredString(250, 157, "Hugging Face")
-
-        self.draw_arrow(canvas, 75, 130, 101, 130, colors.black)
-        self.draw_arrow(canvas, 420, 130, 379, 130, colors.black)
-        self.draw_arrow(canvas, 210, 130, 225, 130, RED)
-        self.draw_arrow(canvas, 290, 130, 275, 130, RED)
-        self.draw_arrow(canvas, 250, 65, 250, 43, colors.black)
-        self.draw_arrow(canvas, 250, 93, 250, 106, RED)
-        self.draw_arrow(canvas, 157, 103, 218, 42, colors.HexColor("#2385CC"))
-        self.draw_arrow(canvas, 332, 103, 282, 42, colors.HexColor("#2385CC"))
-
-        canvas.setFillColor(colors.black)
-        canvas.setFont("DejaVuSans", 6.8)
-        canvas.drawString(5, 106, "individual or organisation")
-        canvas.drawString(402, 106, "individual or organisation")
-        canvas.drawCentredString(250, 3, "individual or organisation")
-        canvas.setFillColor(CNIL_BLUE)
-        canvas.setFont("DejaVuSans-Bold", 13)
-        canvas.drawString(12, 12, "CNIL")
-        canvas.restoreState()
-
-
-def extract_figures(source: Path, work_dir: Path) -> tuple[Path, Path]:
-    reader = PdfReader(str(source))
-    fox_image = reader.pages[2].images[0].image
-    fox_crop = fox_image.crop((0, 0, int(fox_image.width * 0.35), fox_image.height))
-    fox_path = work_dir / "fox.jpg"
-    fox_crop.save(fox_path, quality=92)
-
-    memorisation_path = work_dir / "memorisation.jpg"
-    reader.pages[5].images[0].image.save(memorisation_path, quality=92)
-    return fox_path, memorisation_path
-
-
-def build_styles():
-    pdfmetrics.registerFont(
-        TTFont("DejaVuSans", "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf")
-    )
-    pdfmetrics.registerFont(
-        TTFont(
-            "DejaVuSans-Bold",
-            "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
-        )
-    )
-    styles = getSampleStyleSheet()
-    styles.add(
-        ParagraphStyle(
-            "NoticeTitle",
-            fontName="DejaVuSans-Bold",
-            fontSize=20,
-            leading=24,
-            textColor=CNIL_BLUE,
-            spaceAfter=8 * mm,
-        )
-    )
-    styles.add(
-        ParagraphStyle(
-            "Section",
-            fontName="DejaVuSans-Bold",
-            fontSize=14,
-            leading=17,
-            textColor=CNIL_BLUE,
-            spaceBefore=3 * mm,
-            spaceAfter=2.5 * mm,
-        )
-    )
-    styles.add(
-        ParagraphStyle(
-            "Subsection",
-            fontName="DejaVuSans-Bold",
-            fontSize=11,
-            leading=14,
-            textColor=colors.HexColor("#222222"),
-            spaceBefore=2 * mm,
-            spaceAfter=1.5 * mm,
-        )
-    )
-    styles.add(
-        ParagraphStyle(
-            "BodyNotice",
-            fontName="DejaVuSans",
-            fontSize=8.7,
-            leading=11.3,
-            alignment=4,
-            spaceAfter=2.1 * mm,
-        )
-    )
-    styles.add(
-        ParagraphStyle(
-            "Caption",
-            fontName="DejaVuSans",
-            fontSize=7.2,
-            leading=9,
-            alignment=TA_CENTER,
-            textColor=colors.HexColor("#555555"),
-            spaceAfter=2 * mm,
-        )
-    )
-    return styles
-
-
-def generate(source: Path, output: Path) -> None:
-    styles = build_styles()
-    body = styles["BodyNotice"]
-    title = styles["NoticeTitle"]
-    section = styles["Section"]
-    subsection = styles["Subsection"]
-    caption = styles["Caption"]
-
-    def p(text: str):
-        return Paragraph(text, body)
-
-    def bullet(text: str):
-        style = ParagraphStyle(
-            "BulletNotice",
-            parent=body,
-            leftIndent=5 * mm,
-            firstLineIndent=-3.5 * mm,
-            bulletIndent=1.5 * mm,
-            spaceAfter=1.4 * mm,
-        )
-        return Paragraph(text, style, bulletText="•")
-
-    with tempfile.TemporaryDirectory(prefix="genmod-notice-") as temp_dir:
-        fox_path, memorisation_path = extract_figures(source, Path(temp_dir))
-
-        doc = SimpleDocTemplate(
-            str(output),
-            pagesize=A4,
-            rightMargin=18 * mm,
-            leftMargin=18 * mm,
-            topMargin=18 * mm,
-            bottomMargin=16 * mm,
-            title="A tool for exploring the genealogy of open-source AI models",
-            author="CNIL",
-            subject="English translation of the GenMod information notice",
-        )
-
-        def decorate_page(canvas, document):
-            canvas.saveState()
-            canvas.setStrokeColor(CNIL_BLUE)
-            canvas.setLineWidth(1.2)
-            canvas.line(18 * mm, 12 * mm, A4[0] - 18 * mm, 12 * mm)
-            canvas.setFont("DejaVuSans-Bold", 8)
-            canvas.setFillColor(CNIL_BLUE)
-            canvas.drawString(18 * mm, 7.5 * mm, "CNIL · GenMod")
-            canvas.setFont("DejaVuSans", 8)
-            canvas.drawRightString(
-                A4[0] - 18 * mm, 7.5 * mm, f"Page {document.page}"
-            )
-            canvas.restoreState()
-
-        story = [
-            Paragraph(
-                "A tool for exploring the genealogy of open-source AI models",
-                title,
-            ),
-            Paragraph("What is an AI model?", section),
-            Paragraph("Training", subsection),
-            p(
-                "The fields in which AI can be used are vast and difficult to delimit, as they extend to many aspects of everyday life: online searches and purchases, targeted advertising, machine translation, personal digital assistants and connected cities, as well as transport, healthcare and many other areas."
-            ),
-            p(
-                "Under Article 3 of the European Union Artificial Intelligence Act, an AI system is ‘a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.’"
-            ),
-            p(
-                "These systems incorporate one or more AI models. Such models can be described as algorithms whose operation is determined by a set of attributes and which are designed to perform tasks such as prediction, classification, inference or generation. Deep neural-network models, for example, consist of nodes (neurons) arranged in layers and connected by edges, each of which has a parameter or ‘weight’. During training, these parameters are adjusted to learn the statistical distribution of the training data."
-            ),
-            p("For a simple neural network, the model’s attributes might include:"),
-            bullet("the type and size of each layer (linear, convolutional, attention, etc.);"),
-            bullet("the weights assigned to each edge (also called parameters);"),
-            bullet("the activation functions between layers; and"),
-            bullet("possibly other operations located within or between layers."),
-            Spacer(1, 1.5 * mm),
-            NeuralNetworkDiagram(),
-            Paragraph("Figure 1 — Diagram of a neural network (authors)", caption),
-            PageBreak(),
-            Paragraph("Training from examples", subsection),
-            p(
-                "When a neural network is trained to recognise images, it is given examples in which the image pixels are associated with an annotation, or label. The model then adjusts its parameters—the weights—to learn to assign the correct label as often as possible."
-            ),
-            p(
-                "The main difference between a deep-learning model and a conventional computer program is that the model learns inference rules autonomously from data. In a conventional program, a task is solved using explicit rules defined in advance by the developer. To sort a list of numbers, for example, the order in which elements are compared is precisely programmed. This works very well for clearly delimited tasks for which explicit rules can be established."
-            ),
-            p(
-                "By contrast, the rules of a deep-learning model are not specified directly. Instead, the model is given a large volume of examples—its training data—so that, during the learning phase, it can identify the statistical regularities or strategies that solve the task. This makes it possible to automate much more complex tasks for which defining every rule by hand would be extremely difficult or impossible."
-            ),
-            Paragraph("Using a trained model", subsection),
-            p(
-                "Once trained, a model can be used as it is, without further modification, to perform specific tasks automatically. This is called the inference phase. The model receives an input—an image, text or an audio signal, for example—and produces an output based on what it learned during training. It then acts as a ‘black box’: it applies the regularities it has learned without changing its internal structure or learning anything new."
-            ),
-            p(
-                "Consider machine translation. A neural-network model trained on millions of pairs of Spanish and English sentences can translate a new text from English into Spanish at inference time. Linguistic rules are not explicitly implemented in the model; it has learned to match sequences of words by drawing on statistical regularities in the training data."
-            ),
-            p(
-                "Another example is image-to-text models, which generate image captions. A model can be trained to associate images with textual descriptions. Once trained, it can receive a new image—such as a photograph of a dog running in a park—and automatically produce a sentence such as ‘A dog is running on the grass in a park.’"
-            ),
-            Table(
-                [[
-                    Image(str(fox_path), width=58 * mm, height=41 * mm),
-                    Paragraph(
-                        "The image is a close-up portrait of a red fox standing in the snow. The fox is in the centre, its vibrant orange fur lit by golden sunrise or sunset light. It stands alert, ears upright and looking off-camera. The pristine white snow has a bluish tint reflecting the cool colours of the sky. The softly blurred blue-grey background adds depth and highlights the fox as the subject.",
-                        body,
-                    ),
-                ]],
-                colWidths=[62 * mm, 105 * mm],
-                style=TableStyle([
-                    ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
-                    ("LEFTPADDING", (0, 0), (-1, -1), 2),
-                    ("RIGHTPADDING", (0, 0), (-1, -1), 4),
-                    ("BOX", (0, 0), (-1, -1), 0.4, colors.HexColor("#BBBBBB")),
-                    ("BACKGROUND", (0, 0), (-1, -1), colors.HexColor("#F7F7F7")),
-                ]),
-            ),
-            Paragraph(
-                "Figure 2 — Example of an image’s textual description (source: imagedescriber.online)",
-                caption,
-            ),
-            PageBreak(),
-            p(
-                "These use cases illustrate the power of advances in deep learning to automate complex, often subjective or ambiguous tasks for which writing explicit rules by hand would be difficult or impossible."
-            ),
-            Paragraph("Derivatives: fine-tuning, merging, quantisation, etc.", subsection),
-            p(
-                "To adapt a neural network more closely to a specific task, optimise its performance or reduce its running costs, several transformations can be applied to an initial pre-trained model. Such modifications are very common in the open-source ecosystem. They make it possible to create new models from one or more initial models, sometimes with additional data. Four common transformations between a target model and a source model are:"
-            ),
-            bullet(
-                "<b>Fine-tuning:</b> a general-purpose source model continues training on a specific dataset to improve its performance on a more precise task. A large language model initially trained on freely accessible internet sources may, for example, be fine-tuned on a company’s business data so that it better understands the company’s specialist vocabulary and expressions."
-            ),
-            bullet(
-                "<b>Quantisation:</b> the precision of the source model’s weights is reduced to lower its memory footprint. Weights initially encoded using 32 bits may, for example, be rounded to the nearest value that can be encoded using 16 bits."
-            ),
-            bullet(
-                "<b>Adaptation:</b> the source model is adjusted so that it can be used with limited computing resources—for example, on a mobile phone—most often using a Low-Rank Adaptation (LoRA) technique."
-            ),
-            bullet(
-                "<b>Merging:</b> layers from different models are combined to improve performance. For example, two LLMs, A and B, may both have been trained on general text corpora. Averaging the weights in their twelfth layers and replacing A’s twelfth layer with that average may produce a model C that performs better than either A or B."
-            ),
-            PageBreak(),
-            Paragraph("A platform for open-source AI: Hugging Face", section),
-            p(
-                "To enable AI models to be shared and made available by and for as many people as possible, the Franco-American company Hugging Face, founded in 2016, developed a platform that centralises models and datasets. It also provides software tools for deploying AI models. Today it hosts more open-source models than any other platform—over two million were available in September 2025—and acts as a catalyst for the open-source AI ecosystem."
-            ),
-            p("The following example illustrates how the platform works:"),
-            bullet("User C wants to create a model that automatically detects fraudulent emails."),
-            bullet(
-                "User A has published a natural-language-processing model on Hugging Face—for example, Google’s <i>google/gemma-3-27b-it</i>—and User B has published a dataset containing millions of emails labelled as fraudulent or non-fraudulent."
-            ),
-            bullet(
-                "User C downloads the model and dataset, then trains the model on the dataset to specialise it for classification. Once the resulting classifier performs well, User C can publish it on Hugging Face so that anyone can use it as is or train it again on other datasets to improve its performance."
-            ),
-            HuggingFaceDiagram(),
-            Paragraph("Figure 3 — Example of how Hugging Face is used", caption),
-            p(
-                "In short, Hugging Face provides tools for building, training and deploying deep-learning models based on open-source technologies and code. It also offers a space where researchers, engineers and enthusiasts can exchange ideas, obtain support and contribute to open-source projects."
-            ),
-            Paragraph("Benefits of open-source AI", section),
-            p(
-                "The rise of open-source AI shows that powerful and partly transparent models can compete with proprietary solutions while stimulating collective innovation. BLOOM (176 billion parameters, 2022), developed by the BigScience consortium, illustrates this dynamic: trained in 46 languages, it enabled multilingual conversational assistants to be developed in Africa, Latin America and the Arab world, where commercial models remained poorly adapted to local languages."
-            ),
-            PageBreak(),
-            p(
-                "Similarly, GPT-J and GPT-NeoX (EleutherAI) and Vicuna (LMSYS) provided the basis for open-source projects that enabled universities and start-ups to create specialised chatbots without relying on closed services. These models also made research into bias detection and the robustness of large language models possible."
-            ),
-            p(
-                "In computer vision, Stable Diffusion (Stability AI) transformed visual creation. Its weights were made freely available, opening the way to applications in video games, advertising and audiovisual production, including concept images, storyboards and rapid design. Its open-source code enabled tools such as Automatic1111 and ComfyUI, used by hundreds of thousands of artists and researchers."
-            ),
-            p(
-                "The impact is industrial as well. LLaMA (Meta), initially released to the research community, gave rise to a generation of derivatives—including Zephyr, Nous-Hermes and OpenChat—that are now used for customer support, summarising legal or medical documents and prototyping code."
-            ),
-            p(
-                "In science, projects such as BioGPT (Microsoft Research) and OpenFold (inspired by AlphaFold) demonstrate how opening code and weights accelerates biomedical research by allowing independent laboratories to reproduce and improve results in protein-structure prediction and molecule discovery."
-            ),
-            p(
-                "These achievements show that open source is not limited to reusing models: it enables technological ownership, local adaptation and open innovation in fields as varied as artistic creation, healthcare, education, data science and the cultural industries. Nevertheless, some opacity may remain as to how models are built: with what data and which training algorithm? A CNIL paper and a PEReN paper provide further discussion of this issue."
-            ),
-            Paragraph("Privacy issues", section),
-            Paragraph("Memorisation by AI models", subsection),
-            p(
-                "The scientific community has long established that information about the data used to train an AI model can often be extracted from even partial access to the model. In generative AI, a model may reproduce text or images that are very close to examples in its training dataset. In the figure below, when Stable Diffusion is asked to generate an image matching the caption ‘Emma Watson to play Belle in Disney’s Beauty and the Beast’, its output closely resembles an image from the training database."
-            ),
-            p(
-                "This is regurgitation—only one form of memorisation. Statistical methods can sometimes reveal other information, such as whether a particular record belonged to the training dataset, through membership-inference attacks."
-            ),
-            Image(str(memorisation_path), width=153 * mm, height=91 * mm),
-            Paragraph(
-                "Figure 4 — Source: Louis Hunt (LinkedIn). Original photograph: UN Women.",
-                caption,
-            ),
-            PageBreak(),
-            p(
-                "For text models such as chatbots, prominent cases of regurgitation are already widely documented. One version of ChatGPT, for example, was reported to reproduce New York Times articles almost verbatim and to provide personal information such as a person’s name, address and telephone number."
-            ),
-            Paragraph("The GDPR and AI models", subsection),
-            p(
-                "If information about an AI model’s training database can generally be extracted from the model, what legal regime should apply when that database contains personal data? The European Data Protection Board clarified this question in Opinion 28/2024 on AI models, on which the CNIL’s latest recommendations are based. In particular, the Opinion concludes that the GDPR applies in many cases to AI models trained on personal data because of their capacity for memorisation."
-            ),
-            Paragraph("Exercising rights in relation to AI models", subsection),
-            p(
-                "For AI models subject to the GDPR, people affected by memorisation have rights in relation to their data, including the rights to object, access and erasure. These rights are not absolute: a controller may depart from them in several situations, for example where a request is manifestly unfounded or excessive (Article 12), or where the controller is unable to identify the person concerned. The CNIL’s guidance on exercising rights provides further details."
-            ),
-            p(
-                "At a time when European bodies are confirming that data-protection law also applies to AI models, the CNIL wishes to study the conditions under which these rights could be exercised within the highly dynamic open-source AI ecosystem."
-            ),
-            Spacer(1, 8 * mm),
-            Table(
-                [[Paragraph(
-                    "This document is an English translation of the French information notice made available by the CNIL for the GenMod application. In the event of any discrepancy, the French version is the reference version.",
-                    ParagraphStyle(
-                        "TranslationNote",
-                        parent=body,
-                        textColor=CNIL_BLUE,
-                        backColor=LIGHT_BLUE,
-                        borderPadding=8,
-                    ),
-                )]],
-                colWidths=[170 * mm],
-            ),
-        ]
-
-        doc.build(story, onFirstPage=decorate_page, onLaterPages=decorate_page)
-
-
-def main() -> None:
-    parser = argparse.ArgumentParser()
-    parser.add_argument(
-        "--source",
-        type=Path,
-        default=Path("application_neo4j/static/notice/notice.pdf"),
-    )
-    parser.add_argument(
-        "--output",
-        type=Path,
-        default=Path("application_neo4j/static/notice/notice_en.pdf"),
-    )
-    args = parser.parse_args()
-    args.output.parent.mkdir(parents=True, exist_ok=True)
-    generate(args.source, args.output)
-    print(f"Generated {args.output}")
-
-
-if __name__ == "__main__":
-    main()

application_neo4j/static/notice/notice.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:98c0455e1d8419a184a2471fc69960027976f9cb7d9ee07752516c3b216b4a79
-size 1166198
+oid sha256:dced59e9dc43bd1823fd18cee4005415e375e9f77c7a58bc6ffcf3a48e732a3c
+size 409000

application_neo4j/static/notice/notice_en.pdf

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:abec965848c0fd334609151abd08615455a57872fbd0709961e6c8a8ddb2e93b
-size 383279

application_neo4j/templates/expert.html

@@ -18,18 +18,13 @@
 </head>
 
 <body class="d-flex flex-column min-vh-100">
-    <div id="legacy-browser-warning" class="legacy-browser-warning" role="alert" hidden>{{ t('site.legacy_browser_warning') }}</div>
-    <script src="{{ url_for('static', filename='js/browser_compatibility.js') }}"></script>
     <header class="navbar navbar-expand-lg navbar-light bg-light border-bottom">
     <div class="container">
       <a class="navbar-brand" href="{{ url_for('home', lang=current_lang) }}">
         <span class="fw-bold">{{ t('site.nav_brand_model_expert') }}</span>
         <small class="d-block text-muted">{{ t('site.nav_subtitle') }}</small>
       </a>
-      <div class="navbar-actions d-flex align-items-center justify-content-end flex-wrap gap-2">
-        <a href="{{ url_for('home', lang=current_lang) }}" class="btn btn-sm btn-dark me-3">
-          <i class="bi bi-house-door-fill me-1" aria-hidden="true"></i>{{ t('site.home_button') }}
-        </a>
+      <div class="d-flex gap-2">
         <a href="{{ url_for('set_language_route', lang='fr') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'fr' else 'btn-outline-primary' }}">{{ t('lang.fr') }}</a>
         <a href="{{ url_for('set_language_route', lang='en') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'en' else 'btn-outline-primary' }}">{{ t('lang.en') }}</a>
       </div>

application_neo4j/templates/index.html

@@ -5,7 +5,7 @@
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
     <title>{{ t('site.title_home') }}</title>
     <link rel="stylesheet" href="{{ url_for('static', filename='css/style.css') }}">
-    <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.3/dist/css/bootstrap.min.css" rel="stylesheet" integrity="sha384-QWTKZyjpPEjISv5WaRU9OFeRpok6YctnYmDr5pNlyT2bRjXh0JMhjY6hW+ALEwIH" crossorigin="anonymous">
+    <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/css/bootstrap.min.css" rel="stylesheet">
 
 <style>/* Style simple pour la page d'accueil */
  body {
@@ -116,30 +116,11 @@
       background: #3498db;
       color: #fff;
     }
-
-    @media (max-width: 575.98px) {
-      .lang-switch {
-        position: static;
-        justify-content: flex-end;
-        padding: 12px 15px 0;
-      }
-
-      .welcome-container {
-        padding-top: 24px;
-      }
-
-      .content-card {
-        padding: 22px;
-        text-align: left;
-      }
-    }
     
     </style>
     </head>
 
 <body>
-    <div id="legacy-browser-warning" class="legacy-browser-warning" role="alert" hidden>{{ t('site.legacy_browser_warning') }}</div>
-    <script src="{{ url_for('static', filename='js/browser_compatibility.js') }}"></script>
     <!-- Language selector -->
     <div class="lang-switch">
       <a href="{{ url_for('set_language_route', lang='fr') }}" class="{{ 'active' if current_lang == 'fr' else '' }}">{{ t('lang.fr') }}</a>
@@ -148,7 +129,7 @@
 
     <div class="welcome-container">
       <h1>{{ t('home.welcome_title') }}</h1>
-      <a href="{{ url_for('static', filename='notice/notice_en.pdf' if current_lang == 'en' else 'notice/notice.pdf') }}" class="btn-modern" target="_blank">{{ t('home.more_info') }}</a>
+      <a href="{{ url_for('static', filename='notice/notice.pdf') }}" class="btn-modern" target="_blank">{{ t('home.more_info') }}</a>
     </div>
     <div class="container content-section">
       <div class="row g-4">

application_neo4j/templates/infos.html

@@ -1,43 +1,18 @@
 <!DOCTYPE html>
-<html lang="{{ current_lang }}">
+<html lang="fr">
 <head>
     <meta charset="UTF-8">
-    <meta name="viewport" content="width=device-width, initial-scale=1.0">
-    <title>{{ t('home.more_info') }}</title>
+    <title>Plus d'informations</title>
     <style>
         body, html { margin: 0; padding: 0; height: 100%; overflow: hidden; }
-        body { display: flex; flex-direction: column; font-family: sans-serif; }
-        .legacy-browser-warning { display: none; padding: 0.75rem 1rem; color: #664d03; background: #fff3cd; border-bottom: 1px solid #ffecb5; font-weight: 600; text-align: center; }
-        .toolbar {
-            display: flex;
-            align-items: center;
-            padding: 0.75rem 1rem;
-            background: #f8f9fa;
-            border-bottom: 1px solid #dee2e6;
-        }
-        .home-button {
-            color: #fff;
-            background: #212529;
-            border: 1px solid #212529;
-            border-radius: 0.25rem;
-            padding: 0.375rem 0.75rem;
-            text-decoration: none;
-            font-weight: 600;
-        }
-        .home-button:hover { color: #fff; background: #424649; }
-        .pdf-container { width: 100%; height: calc(100% - 52px); flex: 1; min-height: 0; }
-        .pdf-container iframe { display: block; border: none; width: 100%; height: 100%; }
+        .pdf-container { width: 100%; height: 100%; }
+        .pdf-container iframe { border: none; width: 100%; height: 100%; }
     </style>
 </head>
 <body>
-    <div id="legacy-browser-warning" class="legacy-browser-warning" role="alert" hidden>{{ t('site.legacy_browser_warning') }}</div>
-    <script src="{{ url_for('static', filename='js/browser_compatibility.js') }}"></script>
-    <nav class="toolbar" aria-label="{{ t('site.home_button') }}">
-        <a class="home-button" href="{{ url_for('home', lang=current_lang) }}">⌂ {{ t('site.home_button') }}</a>
-    </nav>
     <div class="pdf-container">
         <!-- Cette balise iframe va afficher votre PDF -->
         <iframe src="{{ url_for('static', filename='pdf/plus_infos_interface.pdf') }}"></iframe>
     </div>
 </body>
-</html>
+</html>

application_neo4j/templates/search.html

@@ -33,18 +33,13 @@
 </head>
 
 <body class="d-flex flex-column min-vh-100">
-    <div id="legacy-browser-warning" class="legacy-browser-warning" role="alert" hidden>{{ t('site.legacy_browser_warning') }}</div>
-    <script src="{{ url_for('static', filename='js/browser_compatibility.js') }}"></script>
     <header class="navbar navbar-expand-lg navbar-light bg-light border-bottom">
     <div class="container">
       <a class="navbar-brand" href="{{ url_for('home', lang=current_lang) }}">
         <span class="fw-bold">{{ t('site.nav_brand') }}</span>
         <small class="d-block text-muted">{{ t('site.nav_subtitle_full') }}</small>
       </a>
-      <div class="navbar-actions d-flex align-items-center justify-content-end flex-wrap gap-2">
-        <a href="{{ url_for('home', lang=current_lang) }}" class="btn btn-sm btn-dark me-3">
-          <i class="bi bi-house-door-fill me-1" aria-hidden="true"></i>{{ t('site.home_button') }}
-        </a>
+      <div class="d-flex gap-2">
         <a href="{{ url_for('set_language_route', lang='fr') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'fr' else 'btn-outline-primary' }}">{{ t('lang.fr') }}</a>
         <a href="{{ url_for('set_language_route', lang='en') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'en' else 'btn-outline-primary' }}">{{ t('lang.en') }}</a>
       </div>
@@ -179,8 +174,8 @@
                                       <span class="badge {{ badge.class }} on-top" 
                                             data-bs-toggle="tooltip" 
                                             data-bs-placement="top" 
-                                            title="{{ t(badge.title_key) }}">
-                                          {{ t(badge.text_key) }}
+                                            title="{{ badge.title }}">
+                                          {{ badge.text }}
                                       </span>
                                   {% endfor %}
                                     </div>
@@ -305,8 +300,8 @@
                           <span class="badge {{ badge.class }} on-top" 
                                 data-bs-toggle="tooltip" 
                                 data-bs-placement="top" 
-                                title="{{ t(badge.title_key) }}">
-                              {{ t(badge.text_key) }}
+                                title="{{ badge.title }}">
+                              {{ badge.text }}
                           </span>
                       {% endfor %}
                         </div>

application_neo4j/templates/search_dataset.html

@@ -33,18 +33,13 @@
 </head>
 
 <body class="d-flex flex-column min-vh-100">
-    <div id="legacy-browser-warning" class="legacy-browser-warning" role="alert" hidden>{{ t('site.legacy_browser_warning') }}</div>
-    <script src="{{ url_for('static', filename='js/browser_compatibility.js') }}"></script>
     <header class="navbar navbar-expand-lg navbar-light bg-light border-bottom">
     <div class="container">
       <a class="navbar-brand" href="{{ url_for('home', lang=current_lang) }}">
         <span class="fw-bold">{{ t('site.nav_brand') }}</span>
         <small class="d-block text-muted">{{ t('site.nav_subtitle') }}</small>
       </a>
-      <div class="navbar-actions d-flex align-items-center justify-content-end flex-wrap gap-2">
-        <a href="{{ url_for('home', lang=current_lang) }}" class="btn btn-sm btn-dark me-3">
-          <i class="bi bi-house-door-fill me-1" aria-hidden="true"></i>{{ t('site.home_button') }}
-        </a>
+      <div class="d-flex gap-2">
         <a href="{{ url_for('set_language_route', lang='fr') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'fr' else 'btn-outline-primary' }}">{{ t('lang.fr') }}</a>
         <a href="{{ url_for('set_language_route', lang='en') }}" class="btn btn-sm {{ 'btn-primary' if current_lang == 'en' else 'btn-outline-primary' }}">{{ t('lang.en') }}</a>
       </div>

application_neo4j/translations.py

@@ -45,8 +45,6 @@ TRANSLATIONS = {
         "site.nav_brand_model_expert": "Généalogie des Modèles",
         "site.nav_subtitle": "Exploration des relations entre modèles et datasets",
         "site.nav_subtitle_full": "Exploration des relations entre modèles et datasets (base de données actualisée le {date})",
-        "site.home_button": "Accueil",
-        "site.legacy_browser_warning": "Ce navigateur n’est plus pris en charge. Pour un affichage stable, utilisez une version récente de Firefox, Chrome, Edge ou Safari.",
         "site.footer": "Application de recherche et visualisation des relations entre modèles et jeux de données publiés sur la plateforme HuggingFace. © 2025",
         "site.footer_expert": "Application de recherche et visualisation... © 2025",
 
@@ -100,7 +98,7 @@ TRANSLATIONS = {
         "search.page_lead_dataset": "Explorez la généalogie des modèles et datasets",
         "search.label_name": "Nom à rechercher",
         "search.placeholder_model": "Taper le nom du modèle ou l'id de son repo Hugging Face.",
-        "search.placeholder_dataset": "Taper le nom du dataset à investiguer.",
+        "search.placeholder_dataset": "Taper le nom du dataset suspecté.",
         "search.label_filters": "Filtres",
         "search.filter_model": "Modèle",
         "search.filter_dataset": "Dataset",
@@ -110,7 +108,7 @@ TRANSLATIONS = {
         "search.btn_search": "Rechercher",
         "search.btn_search_simple": "Rechercher",
         "search.progress_title": "Recherche en cours…",
-        "search.progress_stage_queued": "En attente d’une capacité de recherche…",
+        "search.progress_stage_queued": "En attente du serveur…",
         "search.progress_stage_preparing": "Préparation du graphe…",
         "search.progress_stage_descendants": "Recherche des descendants…",
         "search.progress_stage_ancestors": "Recherche des ascendants…",
@@ -124,14 +122,6 @@ TRANSLATIONS = {
         "search.progress_stage_cancelled": "Recherche annulée.",
         "search.progress_elapsed": "Temps écoulé : {seconds} s",
         "search.progress_items": "Éléments traités par le serveur : {completed}/{total}",
-        "search.progress_queue_title": "Recherche en file d’attente",
-        "search.progress_queue_position": "Position dans la file : {position}/{total}",
-        "search.progress_queue_ahead_one": "1 recherche devant la vôtre",
-        "search.progress_queue_ahead_many": "{count} recherches devant la vôtre",
-        "search.progress_queue_next": "Votre recherche est la prochaine.",
-        "search.progress_queue_running_one": "1 recherche en cours",
-        "search.progress_queue_running_many": "{count} recherches en cours",
-        "search.progress_queue_note": "La recherche démarrera automatiquement dès qu’une capacité sera disponible.",
         "search.progress_server_percent": "Progression indiquée par le serveur : {percent} %",
         "search.progress_remaining_step": "Temps restant estimé pour cette étape : {seconds} s",
         "search.progress_server_note": "La progression repose sur les opérations réellement terminées par le serveur.",
@@ -197,7 +187,7 @@ TRANSLATIONS = {
         # ── expert.html ──
         "expert.page_title": "Recherche Experte",
         "expert.page_lead": "Explorez et filtrez la généalogie des modèles",
-        "expert.placeholder": "Taper le nom du modèle à investiguer.",
+        "expert.placeholder": "Taper le nom du modèle suspecté.",
         "expert.filter_author": "Auteur",
         "expert.connected_component": "Composante connexe du noeud recherché",
         "expert.legend_title": "Légendes",
@@ -284,8 +274,6 @@ TRANSLATIONS = {
         "site.nav_brand_model_expert": "Model Genealogy",
         "site.nav_subtitle": "Exploration of relations between models and datasets",
         "site.nav_subtitle_full": "Exploration of relations between models and datasets (database updated on {date})",
-        "site.home_button": "Home",
-        "site.legacy_browser_warning": "This browser is no longer supported. For a stable display, use a recent version of Firefox, Chrome, Edge or Safari.",
         "site.footer": "Application for searching and visualizing relations between models and datasets published on the HuggingFace platform. © 2025",
         "site.footer_expert": "Application for searching and visualizing... © 2025",
 
@@ -349,7 +337,7 @@ TRANSLATIONS = {
         "search.btn_search": "Search",
         "search.btn_search_simple": "Search",
         "search.progress_title": "Search in progress…",
-        "search.progress_stage_queued": "Waiting for search capacity…",
+        "search.progress_stage_queued": "Waiting for the server…",
         "search.progress_stage_preparing": "Preparing the graph…",
         "search.progress_stage_descendants": "Searching descendants…",
         "search.progress_stage_ancestors": "Searching ancestors…",
@@ -363,14 +351,6 @@ TRANSLATIONS = {
         "search.progress_stage_cancelled": "Search cancelled.",
         "search.progress_elapsed": "Elapsed time: {seconds} s",
         "search.progress_items": "Items processed by the server: {completed}/{total}",
-        "search.progress_queue_title": "Search queued",
-        "search.progress_queue_position": "Queue position: {position}/{total}",
-        "search.progress_queue_ahead_one": "1 search ahead of yours",
-        "search.progress_queue_ahead_many": "{count} searches ahead of yours",
-        "search.progress_queue_next": "Your search is next.",
-        "search.progress_queue_running_one": "1 search currently running",
-        "search.progress_queue_running_many": "{count} searches currently running",
-        "search.progress_queue_note": "Your search will start automatically as soon as capacity becomes available.",
         "search.progress_server_percent": "Progress reported by the server: {percent}%",
         "search.progress_remaining_step": "Estimated time remaining for this step: {seconds} s",
         "search.progress_server_note": "Progress is based on operations actually completed by the server.",

database_refresh/refresh_database.py

@@ -66,36 +66,16 @@ def create_views(
     connection.execute(
         f"""
         CREATE VIEW source_models AS
-        SELECT * EXCLUDE (_dedupe_rank)
-        FROM (
-            SELECT
-                *,
-                row_number() OVER (
-                    PARTITION BY id
-                    ORDER BY lastModified DESC NULLS LAST, _id DESC NULLS LAST
-                ) AS _dedupe_rank
-            FROM read_parquet('{sql_path(models_path)}')
-            WHERE id IS NOT NULL AND trim(id) <> ''
-        )
-        WHERE _dedupe_rank = 1
+        SELECT *
+        FROM read_parquet('{sql_path(models_path)}')
         {model_limit}
         """
     )
     connection.execute(
         f"""
         CREATE VIEW source_datasets AS
-        SELECT * EXCLUDE (_dedupe_rank)
-        FROM (
-            SELECT
-                *,
-                row_number() OVER (
-                    PARTITION BY id
-                    ORDER BY lastModified DESC NULLS LAST, _id DESC NULLS LAST
-                ) AS _dedupe_rank
-            FROM read_parquet('{sql_path(datasets_path)}')
-            WHERE id IS NOT NULL AND trim(id) <> ''
-        )
-        WHERE _dedupe_rank = 1
+        SELECT *
+        FROM read_parquet('{sql_path(datasets_path)}')
         {dataset_limit}
         """
     )
@@ -296,6 +276,19 @@ def header_for(path: Path) -> Path:
 
 
 def build_dump(files: dict[str, Path], output_dir: Path, neo4j_admin: str) -> Path:
+    schema_path = output_dir / "schema.cypher"
+    schema_path.write_text(
+        "\n".join(
+            (
+                "CREATE INDEX model_name IF NOT EXISTS FOR (n:Model) ON (n.name);",
+                "CREATE INDEX dataset_name IF NOT EXISTS FOR (n:Dataset) ON (n.name);",
+                "CREATE INDEX author_name IF NOT EXISTS FOR (n:Author) ON (n.name);",
+                "",
+            )
+        ),
+        encoding="utf-8",
+    )
+
     def group(name: str) -> str:
         return f"{header_for(files[name])},{files[name]}"
 
@@ -308,7 +301,7 @@ def build_dump(files: dict[str, Path], output_dir: Path, neo4j_admin: str) -> Pa
         "--overwrite-destination=true",
         "--id-type=string",
         "--threads=2",
-        "--verbose",
+        f"--schema={schema_path}",
         f"--nodes=Model={group('models')}",
         f"--nodes=Dataset={group('datasets')}",
         f"--nodes=Author={group('authors')}",
@@ -355,14 +348,10 @@ def write_metadata(
     return path
 
 
-def parquet_unique_id_count(path: Path) -> int:
+def parquet_count(path: Path) -> int:
     connection = duckdb.connect()
     count = connection.execute(
-        f"""
-        SELECT count(DISTINCT id)
-        FROM read_parquet('{sql_path(path)}')
-        WHERE id IS NOT NULL AND trim(id) <> ''
-        """
+        f"SELECT count(*) FROM read_parquet('{sql_path(path)}')"
     ).fetchone()[0]
     connection.close()
     return int(count)
@@ -452,8 +441,8 @@ def main() -> int:
         return 0
 
     dump_path = build_dump(files, args.work_dir, args.neo4j_admin)
-    model_count = args.max_models or parquet_unique_id_count(models_path)
-    dataset_count = args.max_datasets or parquet_unique_id_count(datasets_path)
+    model_count = args.max_models or parquet_count(models_path)
+    dataset_count = args.max_datasets or parquet_count(datasets_path)
     metadata_path = write_metadata(
         args.work_dir,
         source_revision,

database_refresh/run_weekly_refresh.sh

@@ -1,7 +1,15 @@
-#!/bin/bash
-set -e
-apt-get update
-apt-get install -y python3-pip wget
-pip3 install duckdb==1.3.2 huggingface-hub==0.31.4
-wget -qO /tmp/r.py https://huggingface.co/spaces/cnil/genmod/resolve/refresh-hub-database/database_refresh/refresh_database.py
-exec python3 /tmp/r.py
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+if ! command -v python3 >/dev/null 2>&1; then
+  apt-get update
+  apt-get install -y python3 python3-pip
+fi
+
+python3 -m pip install --break-system-packages \
+  "duckdb==1.3.2" \
+  "huggingface-hub>=0.31.4,<2"
+
+exec python3 "${SCRIPT_DIR}/refresh_database.py" "$@"

start.sh

@@ -1,7 +1,7 @@
 set -euo pipefail
 
-DUMP_REPO="${GENMOD_DUMP_REPO:-cnil/genmod-dump-neo4j}"
-DUMP_REVISION="${GENMOD_DUMP_REVISION:-main}"
+DUMP_REPO="${NEO4J_DUMP_REPO:-cnil/genmod-dump-neo4j}"
+DUMP_REVISION="${NEO4J_DUMP_REVISION:-main}"
 DUMP_BASE_URL="https://huggingface.co/datasets/${DUMP_REPO}/resolve/${DUMP_REVISION}"
 AUTH_HEADER="Authorization: Bearer ${HF_TOKEN}"
 
@@ -18,20 +18,6 @@ neo4j-admin database load --expand-commands neo4j --from-path=/backups --overwri
 
 # start database
 /startup/docker-entrypoint.sh neo4j &
-
-# The Community importer cannot create indexes through --schema. Ensure that
-# dumps produced by the refresh Job receive the required indexes before the
-# application starts serving searches.
-until cypher-shell -u neo4j -p genealogiemodeles \
-  "RETURN 1;" >/dev/null 2>&1; do
-  sleep 2
-done
-cypher-shell -u neo4j -p genealogiemodeles \
-  "CREATE INDEX IF NOT EXISTS FOR (m:Model) ON (m.name);
-   CREATE INDEX IF NOT EXISTS FOR (a:Author) ON (a.name);
-   CREATE INDEX IF NOT EXISTS FOR (d:Dataset) ON (d.name);
-   CALL db.awaitIndexes(600);"
-
 # start tool
 python3 /application_neo4j/app.py neo4j genealogiemodeles &
 # wait for any process to exit