Add deepseek

app.py

@@ -18,10 +18,12 @@ from transformers import AutoModelForCausalLM
 
 models = {
     #"OS-Copilot/OS-Atlas-Base-7B": Qwen2VLForConditionalGeneration.from_pretrained("OS-Copilot/OS-Atlas-Base-7B", torch_dtype="auto", device_map="auto"),
+    "deepseek-ai/deepseek-vl2-tiny": AutoModelForCausalLM.from_pretrained("deepseek-ai/deepseek-vl2-tiny", trust_remote_code=True)
 }
 
 processors = {
     #"OS-Copilot/OS-Atlas-Base-7B": AutoProcessor.from_pretrained("OS-Copilot/OS-Atlas-Base-7B")
+    "deepseek-ai/deepseek-vl2-tiny": DeepseekVLV2Processor.from_pretrained("deepseek-ai/deepseek-vl2-tiny",)
 }
 
 
@@ -56,7 +58,7 @@ def rescale_bounding_boxes(bounding_boxes, original_width, original_height, scal
     return rescaled_boxes
 
 
-def deepseek():
+def deepseek(image):
     print("helloe!!!!")
     # specify the path to the model
     model_path = "deepseek-ai/deepseek-vl2-tiny"
@@ -80,7 +82,7 @@ def deepseek():
     pil_images = load_pil_images(conversation)
     prepare_inputs = vl_chat_processor(
         conversations=conversation,
-        images=pil_images,
+        images=[image],
         force_batchify=True,
         system_prompt=""
     ).to(vl_gpt.device)
@@ -107,7 +109,7 @@ def deepseek():
 @spaces.GPU
 def run_example(image, text_input, model_id="OS-Copilot/OS-Atlas-Base-7B"):
 
-    deepseek()
+    deepseek(image, )
     
 
 def run_example_old(image, text_input, model_id="OS-Copilot/OS-Atlas-Base-7B"):