Fix paraphrase minilm (#436)

* fix: Fix minilm paraphrase by adding it to pool models * tests: Updated minilm paraphrase canonical vector * chore: Added a warning message for updating the model * chore: Added version where model will be removed
qdrant · Jan 27, 2025 · c2f6fd1 · c2f6fd1
1 parent b05877d
commit c2f6fd1
Show file tree

Hide file tree

Showing 4 changed files with 20 additions and 12 deletions.
diff --git a/fastembed/text/onnx_embedding.py b/fastembed/text/onnx_embedding.py
@@ -79,17 +79,6 @@
         },
         "model_file": "model_optimized.onnx",
     },
-    {
-        "model": "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
-        "dim": 384,
-        "description": "Text embeddings, Unimodal (text), Multilingual (~50 languages), 512 input tokens truncation, Prefixes for queries/documents: not necessary, 2019 year.",
-        "license": "apache-2.0",
-        "size_in_GB": 0.22,
-        "sources": {
-            "hf": "qdrant/paraphrase-multilingual-MiniLM-L12-v2-onnx-Q",
-        },
-        "model_file": "model_optimized.onnx",
-    },
     {
         "model": "thenlper/gte-large",
         "dim": 1024,

diff --git a/fastembed/text/pooled_embedding.py b/fastembed/text/pooled_embedding.py
@@ -40,6 +40,17 @@
         },
         "model_file": "onnx/model.onnx",
     },
+    {
+        "model": "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
+        "dim": 384,
+        "description": "Text embeddings, Unimodal (text), Multilingual (~50 languages), 512 input tokens truncation, Prefixes for queries/documents: not necessary, 2019 year.",
+        "license": "apache-2.0",
+        "size_in_GB": 0.22,
+        "sources": {
+            "hf": "qdrant/paraphrase-multilingual-MiniLM-L12-v2-onnx-Q",
+        },
+        "model_file": "model_optimized.onnx",
+    },
 ]
 
 

diff --git a/fastembed/text/text_embedding.py b/fastembed/text/text_embedding.py
@@ -72,6 +72,14 @@ def __init__(
                 UserWarning,
                 stacklevel=2,
             )
+        if model_name == "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2":
+            warnings.warn(
+                "The model 'sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2' has been updated to "
+                "include a mean pooling layer. Please ensure your usage aligns with the new functionality. "
+                "Support for the previous version without mean pooling will be removed as of version 0.5.2.",
+                UserWarning,
+                stacklevel=2,
+            )
         for EMBEDDING_MODEL_TYPE in self.EMBEDDINGS_REGISTRY:
             supported_models = EMBEDDING_MODEL_TYPE.list_supported_models()
             if any(model_name.lower() == model["model"].lower() for model in supported_models):

diff --git a/tests/test_text_onnx_embeddings.py b/tests/test_text_onnx_embeddings.py
@@ -32,7 +32,7 @@
         [-0.034478, 0.03102, 0.00673, 0.02611, -0.039362]
     ),
     "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2": np.array(
-        [0.0094, 0.0184, 0.0328, 0.0072, -0.0351]
+        [0.0361, 0.1862, 0.2776, 0.2461, -0.1904]
     ),
     "intfloat/multilingual-e5-large": np.array([0.0098, 0.0045, 0.0066, -0.0354, 0.0070]),
     "sentence-transformers/paraphrase-multilingual-mpnet-base-v2": np.array(