Merge branch 'fix-flash-attn-xformers-config'

README.md

@@ -389,6 +389,14 @@ _, _, text_embeddings, image_embeddings = output
 
 </details>
 
+### On CUDA devices
+
+On a CUDA enabled torch environment, the model comes in `torch.bfloat16` 
+precision by default. When running on CUDA, it is recommended to install 
+[FlashAttention](https://github.com/Dao-AILab/flash-attention?tab=readme-ov-file#installation-and-features) 
+and [xFormers](https://github.com/facebookresearch/xformers?tab=readme-ov-file#installing-xformers) 
+to make use of their efficient attention mechanism implementations.
+
 
 ## License
 

config.json

@@ -28,7 +28,7 @@
             "task_instructions": {
                 "retrieval.query": "Represent the query for retrieving evidence documents: "
             },
-            "use_flash_attn": false
+            "use_flash_attn": true
         },
         "hf_model_name_or_path": "jinaai/jina-embeddings-v3",
         "model_type": "jina_clip_text",
@@ -36,6 +36,7 @@
         "proj_bias": false,
         "proj_type": null
     },
+    "torch_dtype": "bfloat16",
     "truncate_dim": null,
     "use_text_flash_attn": null,
     "use_vision_xformers": null,
@@ -59,6 +60,6 @@
         "rope_embeddings": true,
         "subln": true,
         "width": 1024,
-        "x_attention": false
+        "x_attention": true
     }
 }