remove abundant contents

config.json

@@ -135,12 +135,11 @@
     "max_position_embeddings": 8192,
     "mm_hidden_size": 3072,
     "mm_projector_lr": null,
-    "mm_projector_type": "mlp2x_gelu_segtokv9",
+    "mm_projector_type": "mlp2x_gelu",
     "mm_use_im_patch_token": false,
     "mm_use_im_start_end": false,
     "mm_vision_select_feature": "patch",
     "mm_vision_select_layer": -2,
-    "mm_vision_tower": "segtokv9_xxlarge",
     "num_attention_heads": 32,
     "num_hidden_layers": 32,
     "num_key_value_heads": 8,
@@ -149,9 +148,6 @@
     "rms_norm_eps": 1e-05,
     "rope_scaling": null,
     "rope_theta": 500000.0,
-    "segtok_decoder_dim": 512,
-    "segtok_num_queries": 201,
-    "segtok_posembed": "sinusoidal",
     "tie_word_embeddings": false,
     "tokenizer_model_max_length": 4096,
     "tokenizer_padding_side": "right",

modeling_magma.py

@@ -96,7 +96,7 @@ class MagmaMultiModalProjector(nn.Module):
         vision_backbone = config.get('vision_backbone', 'convnextxxlarge')
         vision_backbone_size = vision_backbone.replace('convnext', '')
         projector_type = config.get('mm_projector_type', 'linear')
-        mlp_gelu_match = re.match(r'^mlp(\d+)x_gelu_segtokv(\d+)$', projector_type)
+        mlp_gelu_match = re.match(r'^mlp(\d+)x_gelu$', projector_type)
         if mlp_gelu_match:
             mlp_depth = int(mlp_gelu_match.group(1))
             modules = [nn.Linear(config['mm_hidden_size'], config['hidden_size'])]