Model: Add vision loading support

Adds the ability to load vision parts of text + image models. Requires
an explicit flag in config because there isn't a way to automatically
determine whether the vision tower should be used.

Signed-off-by: kingbri <bdashore3@proton.me>

backends/exllamav2/model.py

@@ -20,6 +20,7 @@ from exllamav2 import (
     ExLlamaV2Cache_TP,
     ExLlamaV2Tokenizer,
     ExLlamaV2Lora,
+    ExLlamaV2VisionTower,
 )
 from exllamav2.generator import (
     ExLlamaV2Sampler,
@@ -28,6 +29,7 @@ from exllamav2.generator import (
 )
 from itertools import zip_longest
 from loguru import logger
+from PIL import Image
 from typing import List, Optional, Union
 
 from ruamel.yaml import YAML
@@ -91,6 +93,10 @@ class ExllamaV2Container:
     autosplit_reserve: List[float] = [96 * 1024**2]
     use_tp: bool = False
 
+    # Vision vars
+    use_vision: bool = False
+    vision_model: Optional[ExLlamaV2VisionTower] = None
+
     # Load state
     model_is_loading: bool = False
     model_loaded: bool = False
@@ -144,6 +150,9 @@ class ExllamaV2Container:
         # Apply a model's config overrides while respecting user settings
         kwargs = await self.set_model_overrides(**kwargs)
 
+        # Set vision state
+        self.use_vision = unwrap(kwargs.get("vision"), True)
+
         # Prepare the draft model config if necessary
         draft_args = unwrap(kwargs.get("draft_model"), {})
         draft_model_name = draft_args.get("draft_model_name")
@@ -608,6 +617,14 @@ class ExllamaV2Container:
             input_ids = torch.zeros((1, self.config.max_input_len), dtype=torch.long)
             self.draft_model.forward(input_ids, cache=self.cache, preprocess_only=True)
 
+        # Load vision tower if it exists
+        if self.use_vision:
+            self.vision_model = ExLlamaV2VisionTower(self.config)
+
+            for value in self.vision_model.load_gen(callback_gen=progress_callback):
+                if value:
+                    yield value
+
         self.model = ExLlamaV2(self.config)
         if not self.quiet:
             logger.info("Loading model: " + self.config.model_dir)

common/config_models.py

@@ -270,6 +270,12 @@ class ModelConfig(BaseConfigModel):
             "NOTE: Only works with chat completion message lists!"
         ),
     )
+    vision: Optional[bool] = Field(
+        False,
+        description=(
+            "Enables vision support if the model supports it. (default: False)"
+        ),
+    )
     num_experts_per_token: Optional[int] = Field(
         None,
         description=(

common/model.py

@@ -33,6 +33,7 @@ class ModelType(Enum):
     MODEL = "model"
     DRAFT = "draft"
     EMBEDDING = "embedding"
+    VISION = "vision"
 
 
 def load_progress(module, modules):
@@ -70,17 +71,26 @@ async def load_model_gen(model_path: pathlib.Path, **kwargs):
     # Create a new container
     container = await ExllamaV2Container.create(model_path.resolve(), False, **kwargs)
 
-    model_type = "draft" if container.draft_config else "model"
+    # Add possible types of models that can be loaded
+    model_type = [ModelType.MODEL]
+
+    if container.use_vision:
+        model_type.insert(0, ModelType.VISION)
+
+    if container.draft_config:
+        model_type.insert(0, ModelType.DRAFT)
+
     load_status = container.load_gen(load_progress, **kwargs)
 
     progress = get_loading_progress_bar()
     progress.start()
 
     try:
+        index = 0
         async for module, modules in load_status:
             if module == 0:
                 loading_task = progress.add_task(
-                    f"[cyan]Loading {model_type} modules", total=modules
+                    f"[cyan]Loading {model_type[index].value} modules", total=modules
                 )
             else:
                 progress.advance(loading_task)
@@ -89,10 +99,10 @@ async def load_model_gen(model_path: pathlib.Path, **kwargs):
 
             if module == modules:
                 # Switch to model progress if the draft model is loaded
-                if model_type == "draft":
-                    model_type = "model"
-                else:
+                if index == len(model_type):
                     progress.stop()
+                else:
+                    index += 1
     finally:
         progress.stop()
 

config_sample.yml

@@ -124,6 +124,9 @@ model:
   # NOTE: Only works with chat completion message lists!
   prompt_template:
 
+  # Enables vision support if the model supports it. (default: False)
+  vision: false
+
   # Number of experts to use per token.
   # Fetched from the model's config.json if empty.
   # NOTE: For MoE models only.

endpoints/core/types/model.py

@@ -107,6 +107,7 @@ class ModelLoadRequest(BaseModel):
     cache_mode: Optional[str] = None
     chunk_size: Optional[int] = None
     prompt_template: Optional[str] = None
+    vision: Optional[bool] = None
     num_experts_per_token: Optional[int] = None
 
     # Non-config arguments