bug fix

router_backend.py

@@ -238,7 +238,7 @@ def build_model(model_id: str, hf_token: str, ablations: List[str], use_cache: b
     if "olmo" in model_id:
         model_config.vocab_size = len(tokenizer)
 
-    model = model_class.from_pretrained(model_path, config=model_config, low_cpu_mem_usage=True, dtype=torch.float16)
+    model = model_class.from_pretrained(model_path, config=model_config, low_cpu_mem_usage=True)
 
     model.to(DEVICE)
     model = model.half()