Add accelerate, remove device_map for CPU inference

app.py

@@ -15,8 +15,8 @@ tokenizer = AutoTokenizer.from_pretrained("allenai/OLMo-2-1124-7B")
 print("Loading model from HuggingFace...")
 model = AutoModelForCausalLM.from_pretrained(
     REPO_ID,
-    torch_dtype=torch.float32,
-    device_map="cpu",
+    
+    
 )
 model.eval()
 print("Model ready")

requirements.txt

@@ -2,3 +2,4 @@ torch
 transformers>=4.45.0
 huggingface_hub
 numpy
+accelerate