Match API | Added logs with CID | also added max-length on cross encoder

Author: snyaggarwal

core/common/search.py

@@ -2,6 +2,7 @@
 import time
 import urllib
 
+from cid.locals import get_cid
 from django.db.models import Case, When, IntegerField
 from elasticsearch_dsl import FacetedSearch, Q
 from pydash import compact, get
@@ -212,9 +213,10 @@ def to_queryset(self, keep_order=True, normalized_score=False, exact_count=True,
         encoder = bool(txt)
         s, hits, total = self.__get_response(exact_count, encoder)
         max_score = hits.max_score or 1
+        cid = get_cid()
         start_time = time.time()
         hits = get_cross_encoder(txt, hits.hits, encoder_model) if encoder else hits.hits
-        print(f"Cross encoder time: {time.time() - start_time}s")
+        print(f"[{cid}] Cross encoder time: {time.time() - start_time} seconds")
         for result in hits:
             _id = get(result, '_id')
             rerank_score = get(result, '_rerank_score')

core/common/utils.py

@@ -962,7 +962,7 @@ def get_embeddings(txt):
 
 def get_encoder(model):
     if model in ENCODERS:
-        return CrossEncoder(model, device="cpu")
+        return CrossEncoder(model, device="cpu", max_length=128)
     return settings.ENCODER
 
 

core/concepts/views.py

@@ -1,5 +1,6 @@
 import time
 
+from cid.locals import get_cid
 from django.conf import settings
 from django.db.models import F
 from django.http import Http404
@@ -831,20 +832,21 @@ def filter_queryset(self, _=None):  # pylint: disable=too-many-locals,too-many-s
         reranker = self.request.GET.get('reranker', None) in get_truthy_values()  # enables reranker
         reranker = reranker and self.request.user.is_mapper_cross_encoder_group
         score_to_sort = 'search_rerank_score' if reranker else 'search_normalized_score'
+        cid = get_cid()
         results = []
         for row in rows:
             start_time = time.time()
             search = ConceptFuzzySearch.search(
                 row, target_repo_url, repo_params, include_retired,
                 is_semantic, num_candidates, k_nearest, map_config, faceted_criterion, locale_filter
             )
-            print(f"ES Search built in {time.time() - start_time} seconds")
+            print(f"[{cid}] ES Search built in {time.time() - start_time} seconds")
             start_time = time.time()
             search = search.params(track_total_hits=False, request_cache=True)
             es_search = CustomESSearch(search[start:end], ConceptDocument)
             name = row.get('name') or row.get('Name') if reranker else None
             es_search.to_queryset(False, True, False, name, encoder_model)
-            print(f"ES Search (including reranker) executed in {time.time() - start_time} seconds")
+            print(f"[{cid}] ES Search (including reranker) executed in {time.time() - start_time} seconds")
             start_time = time.time()
             result = {'row': row, 'results': [], 'map_config': map_config, 'filter': filters}
             for concept in es_search.queryset:
@@ -859,12 +861,12 @@ def filter_queryset(self, _=None):  # pylint: disable=too-many-locals,too-many-s
                     data = serializer(concept, context={'request': self.request}).data
                     data['search_meta']['search_normalized_score'] = normalized_score * 100
                     result['results'].append(data)
-            print(f"Concepts serialized in {time.time() - start_time} seconds")
+            print(f"[{cid}] Concepts serialized in {time.time() - start_time} seconds")
             start_time = time.time()
             if 'results' in result:
                 result['results'] = sorted(
                     result['results'], key=lambda res: get(res, f'search_meta.{score_to_sort}'), reverse=True)
-            print(f"Concepts sorted in {time.time() - start_time} seconds")
+            print(f"[{cid}] Concepts sorted in {time.time() - start_time} seconds")
             results.append(result)
 
         return results

core/settings.py

@@ -619,7 +619,7 @@
     LM_MODEL_NAME = 'all-MiniLM-L6-v2'
     LM = SentenceTransformer(LM_MODEL_NAME)
     if ENV not in ['qa']:
-        ENCODER = CrossEncoder("BAAI/bge-reranker-v2-m3", device="cpu")
+        ENCODER = CrossEncoder("BAAI/bge-reranker-v2-m3", device="cpu", max_length=128)
 
 ANALYTICS_API = os.environ.get('ANALYTICS_API', 'http://host.docker.internal:8002')
 if ANALYTICS_API: