Upload 3 files

first_working_creation_with_custom_encoders.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a018435767944a31ac38c17dfadd98f681af86227195db88c1763dcd3786ee9
+size 2468592399

josie_architecture.txt

@@ -0,0 +1,1002 @@
+JOSIE(
+  (encoder): Encoder(
+    (modality_preprocessors): ModuleDict(
+      (vision): RGBDTPreprocessor(
+        (cls_token): tensor((1, 1, 768), requires_grad=False)
+        
+        (rgbt_stem): PatchEmbedGeneric(
+          (proj): Sequential(
+            (0): PadIm2Video()
+            (1): Conv3d(3, 768, kernel_size=(2, 14, 14), stride=(2, 14, 14), bias=False)
+          )
+        )
+        (pos_embedding_helper): SpatioTemporalPosEmbeddingHelper(
+          (pos_embed): tensor((1, 7681, 768), requires_grad=False)
+          
+        )
+      )
+      (audio): AudioPreprocessor(
+        (cls_token): tensor((1, 1, 768), requires_grad=False)
+        
+        (rgbt_stem): PatchEmbedGeneric(
+          (proj): Conv2d(1, 768, kernel_size=(16, 16), stride=(10, 10), bias=False)
+          (norm_layer): RMSNorm()
+        )
+        (pos_embedding_helper): SpatioTemporalPosEmbeddingHelper(
+          (pos_embed): tensor((1, 229, 768), requires_grad=False)
+          
+        )
+      )
+      (depth): RGBDTPreprocessor(
+        (cls_token): tensor((1, 1, 384), requires_grad=False)
+        
+        (depth_stem): PatchEmbedGeneric(
+          (proj): Conv2d(1, 384, kernel_size=(16, 16), stride=(16, 16), bias=False)
+          (norm_layer): RMSNorm()
+        )
+        (pos_embedding_helper): SpatioTemporalPosEmbeddingHelper(
+          (pos_embed): tensor((1, 197, 384), requires_grad=False)
+          
+        )
+      )
+      (thermal): ThermalPreprocessor(
+        (cls_token): tensor((1, 1, 768), requires_grad=False)
+        
+        (rgbt_stem): PatchEmbedGeneric(
+          (proj): Conv2d(1, 768, kernel_size=(16, 16), stride=(16, 16), bias=False)
+          (norm_layer): RMSNorm()
+        )
+        (pos_embedding_helper): SpatioTemporalPosEmbeddingHelper(
+          (pos_embed): tensor((1, 197, 768), requires_grad=False)
+          
+        )
+      )
+    )
+    (modality_transformers): ModuleDict(
+      (vision): EncoderTransformer(
+        (pre_transformer_layer): Sequential(
+          (0): RMSNorm()
+          (1): EinOpsRearrange()
+        )
+        (post_transformer_layer): EinOpsRearrange()
+        (blocks): ModuleList(
+          (0): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (1): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (2): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (3): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (4): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (5): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (6): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (7): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (8): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (9): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (10): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (11): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+        )
+      )
+      (audio): EncoderTransformer(
+        (pre_transformer_layer): Sequential(
+          (0): RMSNorm()
+          (1): EinOpsRearrange()
+        )
+        (post_transformer_layer): EinOpsRearrange()
+        (blocks): ModuleList(
+          (0): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (1): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.009)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (2): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.018)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (3): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.027)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (4): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.036)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (5): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.045)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (6): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.055)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (7): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.064)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (8): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.073)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (9): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.082)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (10): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.091)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (11): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): DropPath(drop_prob=0.100)
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+        )
+      )
+      (depth): EncoderTransformer(
+        (pre_transformer_layer): Sequential(
+          (0): RMSNorm()
+          (1): EinOpsRearrange()
+        )
+        (post_transformer_layer): EinOpsRearrange()
+        (blocks): ModuleList(
+          (0): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+          (1): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+          (2): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+          (3): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+          (4): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+          (5): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=384, out_features=384, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=384, out_features=256, bias=False)
+              (w2): Linear(in_features=256, out_features=384, bias=False)
+              (w3): Linear(in_features=384, out_features=256, bias=False)
+            )
+          )
+        )
+      )
+      (thermal): EncoderTransformer(
+        (pre_transformer_layer): Sequential(
+          (0): RMSNorm()
+          (1): EinOpsRearrange()
+        )
+        (post_transformer_layer): EinOpsRearrange()
+        (blocks): ModuleList(
+          (0): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (1): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (2): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (3): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (4): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+          (5): EncoderTransformerBlock(
+            (attn): MultiheadAttention(
+              (out_proj): NonDynamicallyQuantizableLinear(in_features=768, out_features=768, bias=True)
+            )
+            (drop_path): Identity()
+            (norm1): RMSNorm()
+            (norm2): RMSNorm()
+            (mlp): MLP(
+              (w1): Linear(in_features=768, out_features=512, bias=False)
+              (w2): Linear(in_features=512, out_features=768, bias=False)
+              (w3): Linear(in_features=768, out_features=512, bias=False)
+            )
+          )
+        )
+      )
+    )
+    (modality_heads): ModuleDict(
+      (vision): Sequential(
+        (0): RMSNorm()
+        (1): SelectElement()
+        (2): Linear(in_features=768, out_features=1024, bias=False)
+      )
+      (audio): Sequential(
+        (0): RMSNorm()
+        (1): SelectElement()
+        (2): Linear(in_features=768, out_features=1024, bias=False)
+      )
+      (depth): Sequential(
+        (0): RMSNorm()
+        (1): SelectElement()
+        (2): Linear(in_features=384, out_features=1024, bias=False)
+      )
+      (thermal): Sequential(
+        (0): RMSNorm()
+        (1): SelectElement()
+        (2): Linear(in_features=768, out_features=1024, bias=False)
+      )
+    )
+  )
+  (reasoner): Qwen2ForCausalLM(
+    (model): Qwen2Model(
+      (embed_tokens): Embedding(151936, 896)
+      (layers): ModuleList(
+        (0): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (1): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (2): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (3): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (4): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (5): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (6): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (7): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (8): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (9): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (10): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (11): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (12): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (13): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (14): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (15): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (16): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (17): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (18): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (19): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (20): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (21): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (22): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+        (23): Qwen2DecoderLayer(
+          (self_attn): Qwen2Attention(
+            (q_proj): Linear(in_features=896, out_features=896, bias=True)
+            (k_proj): Linear(in_features=896, out_features=128, bias=True)
+            (v_proj): Linear(in_features=896, out_features=128, bias=True)
+            (o_proj): Linear(in_features=896, out_features=896, bias=False)
+            (rotary_emb): Qwen2RotaryEmbedding()
+          )
+          (mlp): Qwen2MLP(
+            (gate_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (up_proj): Linear(in_features=896, out_features=4864, bias=False)
+            (down_proj): Linear(in_features=4864, out_features=896, bias=False)
+            (act_fn): SiLU()
+          )
+          (input_layernorm): Qwen2RMSNorm()
+          (post_attention_layernorm): Qwen2RMSNorm()
+        )
+      )
+      (norm): Qwen2RMSNorm()
+    )
+    (lm_head): Linear(in_features=896, out_features=151936, bias=False)
+  )
+  (input_projetor): Linear(in_features=1024, out_features=896, bias=True)
+)

josie_dict.txt

@@ -0,0 +1,693 @@
+Model's state_dict:
+encoder.modality_preprocessors.vision.cls_token          torch.Size([1, 1, 768])
+encoder.modality_preprocessors.vision.rgbt_stem.proj.1.weight    torch.Size([768, 3, 2, 14, 14])
+encoder.modality_preprocessors.vision.pos_embedding_helper.pos_embed     torch.Size([1, 7681, 768])
+encoder.modality_preprocessors.audio.cls_token   torch.Size([1, 1, 768])
+encoder.modality_preprocessors.audio.rgbt_stem.proj.weight       torch.Size([768, 1, 16, 16])
+encoder.modality_preprocessors.audio.rgbt_stem.norm_layer.weight         torch.Size([768])
+encoder.modality_preprocessors.audio.pos_embedding_helper.pos_embed      torch.Size([1, 229, 768])
+encoder.modality_preprocessors.depth.cls_token   torch.Size([1, 1, 384])
+encoder.modality_preprocessors.depth.depth_stem.proj.weight      torch.Size([384, 1, 16, 16])
+encoder.modality_preprocessors.depth.depth_stem.norm_layer.weight        torch.Size([384])
+encoder.modality_preprocessors.depth.pos_embedding_helper.pos_embed      torch.Size([1, 197, 384])
+encoder.modality_preprocessors.thermal.cls_token         torch.Size([1, 1, 768])
+encoder.modality_preprocessors.thermal.rgbt_stem.proj.weight     torch.Size([768, 1, 16, 16])
+encoder.modality_preprocessors.thermal.rgbt_stem.norm_layer.weight       torch.Size([768])
+encoder.modality_preprocessors.thermal.pos_embedding_helper.pos_embed    torch.Size([1, 197, 768])
+encoder.modality_transformers.vision.pre_transformer_layer.0.weight      torch.Size([768])
+encoder.modality_transformers.vision.blocks.0.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.0.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.0.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.0.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.0.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.0.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.0.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.0.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.0.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.1.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.1.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.1.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.1.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.1.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.1.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.1.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.1.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.1.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.2.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.2.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.2.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.2.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.2.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.2.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.2.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.2.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.2.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.3.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.3.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.3.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.3.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.3.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.3.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.3.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.3.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.3.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.4.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.4.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.4.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.4.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.4.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.4.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.4.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.4.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.4.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.5.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.5.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.5.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.5.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.5.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.5.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.5.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.5.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.5.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.6.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.6.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.6.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.6.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.6.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.6.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.6.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.6.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.6.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.7.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.7.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.7.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.7.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.7.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.7.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.7.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.7.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.7.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.8.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.8.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.8.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.8.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.8.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.8.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.8.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.8.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.8.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.9.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.9.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.vision.blocks.9.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.9.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.vision.blocks.9.norm1.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.9.norm2.weight       torch.Size([768])
+encoder.modality_transformers.vision.blocks.9.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.9.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.9.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.10.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.10.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.vision.blocks.10.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.10.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.vision.blocks.10.norm1.weight      torch.Size([768])
+encoder.modality_transformers.vision.blocks.10.norm2.weight      torch.Size([768])
+encoder.modality_transformers.vision.blocks.10.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.10.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.10.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.11.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.vision.blocks.11.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.vision.blocks.11.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.vision.blocks.11.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.vision.blocks.11.norm1.weight      torch.Size([768])
+encoder.modality_transformers.vision.blocks.11.norm2.weight      torch.Size([768])
+encoder.modality_transformers.vision.blocks.11.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.vision.blocks.11.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.vision.blocks.11.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.audio.pre_transformer_layer.0.weight       torch.Size([768])
+encoder.modality_transformers.audio.blocks.0.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.0.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.0.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.0.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.0.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.0.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.0.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.0.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.0.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.0.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.0.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.1.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.1.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.1.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.1.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.1.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.1.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.1.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.1.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.1.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.1.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.1.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.2.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.2.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.2.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.2.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.2.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.2.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.2.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.2.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.2.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.2.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.2.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.3.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.3.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.3.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.3.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.3.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.3.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.3.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.3.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.3.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.3.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.3.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.4.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.4.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.4.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.4.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.4.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.4.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.4.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.4.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.4.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.4.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.4.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.5.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.5.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.5.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.5.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.5.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.5.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.5.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.5.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.5.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.5.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.5.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.6.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.6.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.6.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.6.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.6.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.6.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.6.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.6.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.6.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.6.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.6.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.7.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.7.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.7.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.7.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.7.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.7.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.7.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.7.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.7.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.7.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.7.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.8.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.8.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.8.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.8.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.8.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.8.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.8.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.8.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.8.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.8.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.8.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.9.attn.in_proj_weight         torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.9.attn.in_proj_bias   torch.Size([2304])
+encoder.modality_transformers.audio.blocks.9.attn.bias_k         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.9.attn.bias_v         torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.9.attn.out_proj.weight        torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.9.attn.out_proj.bias          torch.Size([768])
+encoder.modality_transformers.audio.blocks.9.norm1.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.9.norm2.weight        torch.Size([768])
+encoder.modality_transformers.audio.blocks.9.mlp.w1.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.9.mlp.w2.weight       torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.9.mlp.w3.weight       torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.10.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.10.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.audio.blocks.10.attn.bias_k        torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.10.attn.bias_v        torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.10.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.10.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.audio.blocks.10.norm1.weight       torch.Size([768])
+encoder.modality_transformers.audio.blocks.10.norm2.weight       torch.Size([768])
+encoder.modality_transformers.audio.blocks.10.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.10.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.10.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.11.attn.in_proj_weight        torch.Size([2304, 768])
+encoder.modality_transformers.audio.blocks.11.attn.in_proj_bias          torch.Size([2304])
+encoder.modality_transformers.audio.blocks.11.attn.bias_k        torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.11.attn.bias_v        torch.Size([1, 1, 768])
+encoder.modality_transformers.audio.blocks.11.attn.out_proj.weight       torch.Size([768, 768])
+encoder.modality_transformers.audio.blocks.11.attn.out_proj.bias         torch.Size([768])
+encoder.modality_transformers.audio.blocks.11.norm1.weight       torch.Size([768])
+encoder.modality_transformers.audio.blocks.11.norm2.weight       torch.Size([768])
+encoder.modality_transformers.audio.blocks.11.mlp.w1.weight      torch.Size([512, 768])
+encoder.modality_transformers.audio.blocks.11.mlp.w2.weight      torch.Size([768, 512])
+encoder.modality_transformers.audio.blocks.11.mlp.w3.weight      torch.Size([512, 768])
+encoder.modality_transformers.depth.pre_transformer_layer.0.weight       torch.Size([384])
+encoder.modality_transformers.depth.blocks.0.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.0.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.0.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.0.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.0.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.0.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.0.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.0.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.0.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.0.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.0.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.1.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.1.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.1.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.1.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.1.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.1.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.1.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.1.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.1.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.1.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.1.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.2.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.2.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.2.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.2.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.2.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.2.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.2.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.2.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.2.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.2.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.2.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.3.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.3.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.3.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.3.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.3.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.3.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.3.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.3.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.3.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.3.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.3.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.4.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.4.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.4.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.4.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.4.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.4.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.4.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.4.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.4.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.4.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.4.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.5.attn.in_proj_weight         torch.Size([1152, 384])
+encoder.modality_transformers.depth.blocks.5.attn.in_proj_bias   torch.Size([1152])
+encoder.modality_transformers.depth.blocks.5.attn.bias_k         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.5.attn.bias_v         torch.Size([1, 1, 384])
+encoder.modality_transformers.depth.blocks.5.attn.out_proj.weight        torch.Size([384, 384])
+encoder.modality_transformers.depth.blocks.5.attn.out_proj.bias          torch.Size([384])
+encoder.modality_transformers.depth.blocks.5.norm1.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.5.norm2.weight        torch.Size([384])
+encoder.modality_transformers.depth.blocks.5.mlp.w1.weight       torch.Size([256, 384])
+encoder.modality_transformers.depth.blocks.5.mlp.w2.weight       torch.Size([384, 256])
+encoder.modality_transformers.depth.blocks.5.mlp.w3.weight       torch.Size([256, 384])
+encoder.modality_transformers.thermal.pre_transformer_layer.0.weight     torch.Size([768])
+encoder.modality_transformers.thermal.blocks.0.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.0.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.0.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.0.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.0.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.0.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.0.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.0.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.0.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.0.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.0.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.1.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.1.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.1.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.1.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.1.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.1.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.1.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.1.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.1.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.1.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.1.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.2.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.2.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.2.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.2.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.2.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.2.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.2.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.2.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.2.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.2.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.2.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.3.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.3.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.3.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.3.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.3.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.3.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.3.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.3.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.3.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.3.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.3.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.4.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.4.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.4.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.4.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.4.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.4.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.4.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.4.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.4.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.4.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.4.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.5.attn.in_proj_weight       torch.Size([2304, 768])
+encoder.modality_transformers.thermal.blocks.5.attn.in_proj_bias         torch.Size([2304])
+encoder.modality_transformers.thermal.blocks.5.attn.bias_k       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.5.attn.bias_v       torch.Size([1, 1, 768])
+encoder.modality_transformers.thermal.blocks.5.attn.out_proj.weight      torch.Size([768, 768])
+encoder.modality_transformers.thermal.blocks.5.attn.out_proj.bias        torch.Size([768])
+encoder.modality_transformers.thermal.blocks.5.norm1.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.5.norm2.weight      torch.Size([768])
+encoder.modality_transformers.thermal.blocks.5.mlp.w1.weight     torch.Size([512, 768])
+encoder.modality_transformers.thermal.blocks.5.mlp.w2.weight     torch.Size([768, 512])
+encoder.modality_transformers.thermal.blocks.5.mlp.w3.weight     torch.Size([512, 768])
+encoder.modality_heads.vision.0.weight   torch.Size([768])
+encoder.modality_heads.vision.2.weight   torch.Size([1024, 768])
+encoder.modality_heads.audio.0.weight    torch.Size([768])
+encoder.modality_heads.audio.2.weight    torch.Size([1024, 768])
+encoder.modality_heads.depth.0.weight    torch.Size([384])
+encoder.modality_heads.depth.2.weight    torch.Size([1024, 384])
+encoder.modality_heads.thermal.0.weight          torch.Size([768])
+encoder.modality_heads.thermal.2.weight          torch.Size([1024, 768])
+reasoner.model.embed_tokens.weight       torch.Size([151936, 896])
+reasoner.model.layers.0.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.0.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.0.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.0.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.0.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.0.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.0.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.0.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.0.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.0.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.0.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.0.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.1.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.1.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.1.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.1.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.1.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.1.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.1.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.1.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.1.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.1.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.1.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.1.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.2.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.2.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.2.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.2.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.2.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.2.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.2.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.2.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.2.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.2.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.2.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.2.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.3.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.3.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.3.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.3.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.3.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.3.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.3.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.3.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.3.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.3.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.3.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.3.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.4.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.4.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.4.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.4.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.4.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.4.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.4.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.4.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.4.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.4.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.4.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.4.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.5.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.5.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.5.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.5.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.5.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.5.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.5.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.5.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.5.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.5.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.5.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.5.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.6.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.6.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.6.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.6.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.6.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.6.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.6.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.6.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.6.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.6.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.6.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.6.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.7.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.7.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.7.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.7.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.7.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.7.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.7.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.7.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.7.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.7.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.7.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.7.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.8.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.8.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.8.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.8.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.8.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.8.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.8.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.8.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.8.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.8.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.8.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.8.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.9.self_attn.q_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.9.self_attn.q_proj.bias    torch.Size([896])
+reasoner.model.layers.9.self_attn.k_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.9.self_attn.k_proj.bias    torch.Size([128])
+reasoner.model.layers.9.self_attn.v_proj.weight          torch.Size([128, 896])
+reasoner.model.layers.9.self_attn.v_proj.bias    torch.Size([128])
+reasoner.model.layers.9.self_attn.o_proj.weight          torch.Size([896, 896])
+reasoner.model.layers.9.mlp.gate_proj.weight     torch.Size([4864, 896])
+reasoner.model.layers.9.mlp.up_proj.weight       torch.Size([4864, 896])
+reasoner.model.layers.9.mlp.down_proj.weight     torch.Size([896, 4864])
+reasoner.model.layers.9.input_layernorm.weight   torch.Size([896])
+reasoner.model.layers.9.post_attention_layernorm.weight          torch.Size([896])
+reasoner.model.layers.10.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.10.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.10.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.10.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.10.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.10.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.10.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.10.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.10.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.10.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.10.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.10.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.11.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.11.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.11.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.11.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.11.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.11.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.11.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.11.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.11.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.11.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.11.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.11.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.12.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.12.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.12.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.12.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.12.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.12.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.12.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.12.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.12.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.12.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.12.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.12.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.13.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.13.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.13.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.13.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.13.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.13.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.13.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.13.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.13.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.13.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.13.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.13.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.14.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.14.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.14.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.14.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.14.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.14.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.14.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.14.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.14.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.14.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.14.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.14.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.15.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.15.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.15.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.15.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.15.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.15.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.15.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.15.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.15.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.15.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.15.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.15.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.16.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.16.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.16.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.16.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.16.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.16.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.16.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.16.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.16.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.16.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.16.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.16.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.17.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.17.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.17.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.17.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.17.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.17.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.17.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.17.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.17.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.17.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.17.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.17.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.18.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.18.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.18.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.18.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.18.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.18.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.18.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.18.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.18.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.18.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.18.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.18.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.19.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.19.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.19.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.19.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.19.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.19.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.19.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.19.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.19.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.19.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.19.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.19.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.20.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.20.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.20.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.20.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.20.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.20.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.20.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.20.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.20.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.20.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.20.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.20.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.21.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.21.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.21.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.21.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.21.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.21.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.21.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.21.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.21.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.21.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.21.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.21.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.22.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.22.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.22.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.22.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.22.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.22.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.22.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.22.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.22.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.22.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.22.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.22.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.layers.23.self_attn.q_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.23.self_attn.q_proj.bias   torch.Size([896])
+reasoner.model.layers.23.self_attn.k_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.23.self_attn.k_proj.bias   torch.Size([128])
+reasoner.model.layers.23.self_attn.v_proj.weight         torch.Size([128, 896])
+reasoner.model.layers.23.self_attn.v_proj.bias   torch.Size([128])
+reasoner.model.layers.23.self_attn.o_proj.weight         torch.Size([896, 896])
+reasoner.model.layers.23.mlp.gate_proj.weight    torch.Size([4864, 896])
+reasoner.model.layers.23.mlp.up_proj.weight      torch.Size([4864, 896])
+reasoner.model.layers.23.mlp.down_proj.weight    torch.Size([896, 4864])
+reasoner.model.layers.23.input_layernorm.weight          torch.Size([896])
+reasoner.model.layers.23.post_attention_layernorm.weight         torch.Size([896])
+reasoner.model.norm.weight       torch.Size([896])
+reasoner.lm_head.weight          torch.Size([151936, 896])
+input_projetor.weight    torch.Size([896, 1024])
+input_projetor.bias      torch.Size([896])