feat: reduced default noise of the model

configuration_jina_embeddings_v4.py

@@ -2,6 +2,7 @@ from transformers.models.qwen2_5_vl import Qwen2_5_VLConfig
 
 from typing import Optional
 
+
 class JinaEmbeddingsV4Config(Qwen2_5_VLConfig):
     """
     Configuration for the JinaEmbeddingsV4 model.
@@ -12,10 +13,11 @@ class JinaEmbeddingsV4Config(Qwen2_5_VLConfig):
         single_vector_pool_strategy: str = "mean",
         multi_vector_projector_dim: int = 128,
         pretrained_peft_model_name_or_path: Optional[str] = None,
+        verbosity: int = 0,
         **kwargs,
     ):
         super().__init__(**kwargs)
         self.single_vector_pool_strategy = single_vector_pool_strategy
         self.multi_vector_projector_dim = multi_vector_projector_dim
         self.pretrained_peft_model_name_or_path = pretrained_peft_model_name_or_path
-
+        self.verbosity = verbosity

modeling_jina_embeddings_v4.py

@@ -146,6 +146,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
             self.name_or_path, trust_remote_code=True, use_fast=True
         )
         self.multi_vector_projector_dim = config.multi_vector_projector_dim
+        self.verbosity = config.verbosity
         self._task = None
 
     @property
@@ -335,7 +336,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
             assert not return_numpy, "`return_numpy` is not supported when `return_multivector=True` and more than one data is encoded"
         results = []
         self.eval()
-        for batch in tqdm(dataloader, desc=desc):
+        for batch in tqdm(dataloader, desc=desc, disable=self.verbosity == 0):
             with torch.no_grad():
                 batch = {k: v.to(self.device) for k, v in batch.items()}
                 with torch.autocast(
@@ -349,7 +350,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
                             embeddings = torch.nn.functional.normalize(embeddings, p=2, dim=-1)
                     else:
                         embeddings = embeddings.multi_vec_emb
-                    
+
                     if return_multivector and not return_numpy:
                         valid_tokens = batch["attention_mask"].bool()
                         embeddings = [
@@ -453,7 +454,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
             if return_numpy:
                 print("Warning: `return_numpy` is ignored when `return_multivector=True` and `len(texts) > 1`")
             return_numpy = False
-        
+
         if isinstance(texts, str):
             texts = [texts]
 
@@ -468,7 +469,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
             **encode_kwargs,
         )
 
-        return embeddings if return_list else embeddings[0] 
+        return embeddings if return_list else embeddings[0]
 
     def _load_images_if_needed(
         self, images: List[Union[str, Image.Image]]
@@ -515,7 +516,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
             )
         encode_kwargs = self._validate_encoding_params(truncate_dim=truncate_dim)
         task = self._validate_task(task)
-        
+
         return_list = isinstance(images, list)
 
         # If return_multivector is True and encoding multiple images, ignore return_numpy
@@ -527,7 +528,7 @@ class JinaEmbeddingsV4Model(Qwen2_5_VLForConditionalGeneration):
         # Convert single image to list
         if isinstance(images, (str, Image.Image)):
             images = [images]
-            
+
         images = self._load_images_if_needed(images)
         embeddings = self._process_batches(
             data=images,