Use distilbert

backend/config.py

@@ -10,5 +10,5 @@ QA_MODELS_ID = dict(
 )
 
 SEARCH_MODELS_ID = dict(
-    mpnet_qa='flax-sentence-embeddings/mpnet_stackexchange_v1'
+    distilbert_qa = 'flax-sentence-embeddings/multi-qa_v1-distilbert-cls_dot'
 )

backend/inference.py

@@ -47,7 +47,7 @@ def text_similarity(anchor: str, inputs: List[str], model_name: str, model_dict:
 def text_search(anchor: str, n_answers: int, model_name: str, model_dict: dict):
     # Proceeding with model
     print(model_name)
-    assert model_name == "mpnet_qa"
+    assert model_name == "distilbert_qa"
     model = load_model(model_name, model_dict)
 
     # Creating embeddings
@@ -77,7 +77,7 @@ def text_search(anchor: str, n_answers: int, model_name: str, model_dict: dict):
 def text_cluster(anchor: str, n_answers: int, model_name: str, model_dict: dict):
     # Proceeding with model
     print(model_name)
-    assert model_name == "mpnet_qa"
+    assert model_name == "distilbert_qa"
     model = load_model(model_name, model_dict)
 
     # Creating embeddings

backend/utils.py

@@ -23,7 +23,7 @@ def load_model(model_name, model_dict):
 @st.cache(allow_output_mutation=True)
 def load_embeddings():
     # embedding pre-generated
-    corpus_emb = torch.from_numpy(np.loadtxt('./data/stackoverflow-titles-mpnet-emb.csv', max_rows=10000))
+    corpus_emb = torch.from_numpy(np.loadtxt('./data/stackoverflow-titles-distilbert-emb.csv', max_rows=10000))
     return corpus_emb.float()
 
 @st.cache(allow_output_mutation=True)

data/stackoverflow-titles-distilbert-emb.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f54b58e7835fac510ef46b8ba38c58c9942d769cace977e42a3bb274344ee9f
+size 3916646328