Support llava clip vision model.

comfy/clip_model.py

@@ -211,6 +211,15 @@ def forward(self, pixel_values, attention_mask=None, intermediate_output=None):
             pooled_output = self.post_layernorm(x[:, 0, :])
         return x, i, pooled_output
 
+class LlavaProjector(torch.nn.Module):
+    def __init__(self, in_dim, out_dim, dtype, device, operations):
+        super().__init__()
+        self.linear_1 = operations.Linear(in_dim, out_dim, bias=True, device=device, dtype=dtype)
+        self.linear_2 = operations.Linear(out_dim, out_dim, bias=True, device=device, dtype=dtype)
+
+    def forward(self, x):
+        return self.linear_2(torch.nn.functional.gelu(self.linear_1(x[:, 1:])))
+
 class CLIPVisionModelProjection(torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
@@ -220,7 +229,16 @@ def __init__(self, config_dict, dtype, device, operations):
         else:
             self.visual_projection = lambda a: a
 
+        if "llava3" == config_dict.get("projector_type", None):
+            self.multi_modal_projector = LlavaProjector(config_dict["hidden_size"], 4096, dtype, device, operations)
+        else:
+            self.multi_modal_projector = None
+
     def forward(self, *args, **kwargs):
         x = self.vision_model(*args, **kwargs)
         out = self.visual_projection(x[2])
-        return (x[0], x[1], out)
+        projected = None
+        if self.multi_modal_projector is not None:
+            projected = self.multi_modal_projector(x[1])
+
+        return (x[0], x[1], out, projected)

comfy/clip_vision.py

@@ -65,6 +65,7 @@ def encode_image(self, image, crop=True):
         outputs["last_hidden_state"] = out[0].to(comfy.model_management.intermediate_device())
         outputs["image_embeds"] = out[2].to(comfy.model_management.intermediate_device())
         outputs["penultimate_hidden_states"] = out[1].to(comfy.model_management.intermediate_device())
+        outputs["mm_projected"] = out[3]
         return outputs
 
 def convert_to_transformers(sd, prefix):
@@ -104,7 +105,10 @@ def load_clipvision_from_sd(sd, prefix="", convert_keys=False):
         if sd["vision_model.encoder.layers.0.layer_norm1.weight"].shape[0] == 1152:
             json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_vision_siglip_384.json")
         elif sd["vision_model.embeddings.position_embedding.weight"].shape[0] == 577:
-            json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_vision_config_vitl_336.json")
+            if "multi_modal_projector.linear_1.bias" in sd:
+                json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_vision_config_vitl_336_llava.json")
+            else:
+                json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_vision_config_vitl_336.json")
         else:
             json_config = os.path.join(os.path.dirname(os.path.realpath(__file__)), "clip_vision_config_vitl.json")
     else:

comfy/clip_vision_config_vitl_336_llava.json

@@ -0,0 +1,19 @@
+{
+  "attention_dropout": 0.0,
+  "dropout": 0.0,
+  "hidden_act": "quick_gelu",
+  "hidden_size": 1024,
+  "image_size": 336,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-5,
+  "model_type": "clip_vision_model",
+  "num_attention_heads": 16,
+  "num_channels": 3,
+  "num_hidden_layers": 24,
+  "patch_size": 14,
+  "projection_dim": 768,
+  "projector_type": "llava3",
+  "torch_dtype": "float32"
+}

comfy/sd1_clip.py

@@ -196,8 +196,25 @@ def process_tokens(self, tokens, device):
             index = 0
             pad_extra = 0
             for o in other_embeds:
+                emb = o[1]
+                if torch.is_tensor(emb):
+                    emb = {"type": "embedding", "data": emb}
+
+                emb_type = emb.get("type", None)
+                if emb_type == "embedding":
+                    emb = emb.get("data", None)
+                else:
+                    if hasattr(self.transformer, "preprocess_embed"):
+                        emb = self.transformer.preprocess_embed(emb, device=device)
+                    else:
+                        emb = None
+
+                if emb is None:
+                    index += -1
+                    continue
+
                 ind = index + o[0]
-                emb = o[1].view(1, -1, o[1].shape[-1]).to(device=device, dtype=torch.float32)
+                emb = emb.view(1, -1, emb.shape[-1]).to(device=device, dtype=torch.float32)
                 emb_shape = emb.shape[1]
                 if emb.shape[-1] == tokens_embed.shape[-1]:
                     tokens_embed = torch.cat([tokens_embed[:, :ind], emb, tokens_embed[:, ind:]], dim=1)