Add Gemini OCR and fix VRC log world handling

src/vision/vision_actions.py

@@ -26,18 +26,18 @@ def getVisionPipeline():
 
     try:
         from transformers import AutoProcessor
-        from transformers import AutoModelForImageTextToText
+        from transformers import Gemma4ForConditionalGeneration
         import torch
     except Exception as e:
         raise RuntimeError(f"vision dependencies are missing detail={e}")
 
     log("INFO", f"Vision model initializing model={model_name}")
     processor = AutoProcessor.from_pretrained(model_name)
-    model = AutoModelForImageTextToText.from_pretrained(
+    model = Gemma4ForConditionalGeneration.from_pretrained(
         model_name,
-        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+        torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
         device_map="auto",
-    )
+    ).eval()
     _vision_pipeline = (processor, model)
     return _vision_pipeline
 
@@ -55,6 +55,12 @@ def runVisionFromImage(image_path):
 
     prompt = config["prompt"]
     messages = [
+        {
+            "role": "system",
+            "content": [
+                {"type": "text", "text": "You are a helpful vision-language assistant."},
+            ],
+        },
         {
             "role": "user",
             "content": [
@@ -73,15 +79,23 @@ def runVisionFromImage(image_path):
 
     try:
         image = Image.open(image_path).convert("RGB")
-        inputs = processor(images=image, text=prompt, return_tensors="pt")
-        inputs = {key: value.to(model.device) for key, value in inputs.items()}
+        inputs = processor.apply_chat_template(
+            messages,
+            add_generation_prompt=True,
+            tokenize=True,
+            return_dict=True,
+            return_tensors="pt",
+        )
+        inputs = inputs.to(model.device, dtype=model.dtype)
         with torch.no_grad():
             output = model.generate(
                 **inputs,
                 max_new_tokens=config["max_new_tokens"],
-                temperature=config["temperature"],
+                do_sample=config["temperature"] > 0,
             )
-        text = processor.batch_decode(output, skip_special_tokens=True)[0].strip()
+        input_len = inputs["input_ids"].shape[-1]
+        generated = output[0][input_len:]
+        text = processor.decode(generated, skip_special_tokens=True).strip()
     except Exception as e:
         log("ERROR", f"Vision inference failed detail={e}")
         return None