commit transformers bump

michaelfeil · Jun 9, 2024 · a2435a6 · a2435a6
1 parent 2e1e808
commit a2435a6
Show file tree

Hide file tree

Showing 5 changed files with 11 additions and 9 deletions.
diff --git a/libs/infinity_emb/infinity_emb/transformer/embedder/sentence_transformer.py b/libs/infinity_emb/infinity_emb/transformer/embedder/sentence_transformer.py
@@ -81,7 +81,7 @@ def __init__(self, *, engine_args=EngineArgs):
             )
 
         if engine_args.compile:
-            logger.info("using torch.compile()")
+            logger.info("using torch.compile(dynamic=True)")
             fm.auto_model = torch.compile(fm.auto_model, dynamic=True)
 
     def encode_pre(self, sentences) -> dict[str, "Tensor"]:

diff --git a/libs/infinity_emb/infinity_emb/transformer/vision/torch_vision.py b/libs/infinity_emb/infinity_emb/transformer/vision/torch_vision.py
@@ -40,8 +40,8 @@ def __init__(self, *, engine_args: EngineArgs):
         self.engine_args = engine_args
 
         if engine_args.compile:
-            self.model.vision_model = torch.compile(self.model.vision_model)
-            self.model.text_model = torch.compile(self.model.text_model)
+            self.model.vision_model = torch.compile(self.model.vision_model, dynamic=True)
+            self.model.text_model = torch.compile(self.model.text_model, dynamic=True)
 
         assert hasattr(
             self.model, "get_text_features"

diff --git a/libs/infinity_emb/poetry.lock b/libs/infinity_emb/poetry.lock
diff --git a/libs/infinity_emb/pyproject.toml b/libs/infinity_emb/pyproject.toml
@@ -24,7 +24,7 @@ pydantic = {version = ">=2.4.0,<3", optional=true}
 # backend
 torch = {version = "2.3.1", source = "pypi", optional=true}
 sentence-transformers = {version = "^3.0.1", optional=true} 
-transformers = {version = ">4.34.0,<4.39.2", optional=true} 
+transformers = {version = ">4.34.0,<4.39.4", optional=true} 
 ctranslate2 = {version = "^4.0.0", optional=true}
 optimum = {version = ">=1.16.2", optional=true, extras=["onnxruntime"]}
 hf_transfer = {version=">=0.1.5"}

diff --git a/libs/infinity_emb/tests/end_to_end/test_ct2_sentence.py b/libs/infinity_emb/tests/end_to_end/test_ct2_sentence.py
@@ -52,7 +52,9 @@ def test_load_model(model_base):
     s = ["This is a test sentence."]
     e1 = model_base.encode(s)
     e2 = CT2SentenceTransformer(
-        engine_args=EngineArgs(model_name_or_path=MODEL, device="cpu")
+        engine_args=EngineArgs(
+            model_name_or_path=MODEL, device="cpu", bettertransformer=False
+        )
     ).encode(s)
     np.testing.assert_almost_equal(e1, e2, decimal=6)