fix: cuda error

app.py

@@ -21,9 +21,8 @@ def load_models():
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     model = AutoModelForSeq2SeqLM.from_pretrained(
         model_name,
-        device_map="auto",
-        load_in_8bit=True,
-        trust_remote_code=True
+        torch_dtype=torch.float16,
+        low_cpu_mem_usage=True
     )
     return tokenizer, model
 

requirements.txt

@@ -6,5 +6,4 @@ faiss-cpu
 arxiv
 --extra-index-url https://download.pytorch.org/whl/cpu
 torch
-accelerate>=0.26.0
-bitsandbytes>=0.41.1
+accelerate>=0.26.0