Update app.py

app.py

@@ -10,7 +10,7 @@ import spaces
 import subprocess
 subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
-torch.set_default_device('cuda')
+# torch.set_default_device('cuda')
 
 tokenizer = AutoTokenizer.from_pretrained(
     'qnguyen3/nanoLLaVA',
@@ -22,6 +22,8 @@ model = AutoModelForCausalLM.from_pretrained(
     device_map='auto',
     trust_remote_code=True)
 
+model.to("cuda:0")
+
 class KeywordsStoppingCriteria(StoppingCriteria):
     def __init__(self, keywords, tokenizer, input_ids):
         self.keywords = keywords
@@ -93,14 +95,14 @@ def bot_streaming(message, history):
         tokenize=False,
         add_generation_prompt=True)
     text_chunks = [tokenizer(chunk).input_ids for chunk in text.split('<image>')]
-    input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0)
+    input_ids = torch.tensor(text_chunks[0] + [-200] + text_chunks[1], dtype=torch.long).unsqueeze(0).to("cuda:0")
     stop_str = '<|im_end|>'
     keywords = [stop_str]
     stopping_criteria = KeywordsStoppingCriteria(keywords, tokenizer, input_ids)
     streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
     
-    image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
-    generation_kwargs = dict(input_ids=input_ids.to("cuda:0"), images=image_tensor, streamer=streamer, max_new_tokens=100, stopping_criteria=[stopping_criteria])
+    image_tensor = model.process_images([image], model.config).to("cuda:0")
+    generation_kwargs = dict(input_ids=input_ids, images=image_tensor, streamer=streamer, max_new_tokens=100, stopping_criteria=[stopping_criteria])
     generated_text = ""
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()