Remove unused MultimodalCfg.attn_pooler_heads

README.md

@@ -316,8 +316,7 @@ Training [CoCa](https://arxiv.org/abs/2205.01917) models is enabled through spec
 	"width": 512,
 	"heads": 8,
 	"layers": 12,
-	"latent_dim": 512,
-	"attn_pooler_heads": 8
+	"latent_dim": 512
 }
 ```
 Credit to [lucidrains](https://github.com/lucidrains) for [initial code](https://github.com/lucidrains/CoCa-pytorch), [gpucce](https://github.com/gpucce) for adapting the code to open_clip, and [iejMac](https://github.com/iejMac) for training the models.

src/open_clip/coca_model.py

@@ -47,7 +47,6 @@ class MultimodalCfg(CLIPTextCfg):
     dim_head: int = 64
     heads: int = 8
     n_queries: int = 256
-    attn_pooler_heads: int = 8
 
 
 def _build_text_decoder_tower(

src/open_clip/model_configs/coca_ViT-B-32.json

@@ -23,8 +23,7 @@
         "vocab_size": 49408,
         "width": 512,
         "heads": 8,
-        "layers": 12,
-        "attn_pooler_heads": 8
+        "layers": 12
     },
     "custom_text": true
-}
+}

src/open_clip/model_configs/coca_ViT-L-14.json

@@ -23,8 +23,7 @@
         "vocab_size": 49408,
         "width": 768,
         "heads": 12,
-        "layers": 12,
-        "attn_pooler_heads": 12
+        "layers": 12
     },
     "custom_text": true
 }

src/open_clip/model_configs/coca_base.json

@@ -24,8 +24,7 @@
         "layers": 12,
         "heads": 12,
         "width": 768,
-        "embed_cls": true,
-        "output_tokens": true
+        "embed_cls": true
     },
     "custom_text": true
-}
+}