diff --git a/modeling/inference_models/gptq_hf_torch/class.py b/modeling/inference_models/gptq_hf_torch/class.py
index b44fcd7a..eb3d2475 100644
--- a/modeling/inference_models/gptq_hf_torch/class.py
+++ b/modeling/inference_models/gptq_hf_torch/class.py
@@ -350,7 +350,7 @@ class model_backend(HFTorchInferenceModel):
                 auto_gptq.modeling._base.AutoConfig = hf_bleeding_edge.AutoConfig
                 auto_gptq.modeling._base.AutoModelForCausalLM = hf_bleeding_edge.AutoModelForCausalLM
 
-                model = AutoGPTQForCausalLM.from_quantized(location, model_basename=Path(gptq_file).stem, use_safetensors=gptq_file.endswith(".safetensors"))
+                model = AutoGPTQForCausalLM.from_quantized(location, model_basename=Path(gptq_file).stem, use_safetensors=gptq_file.endswith(".safetensors"), device_map=device_map)
 
                 # Patch in embeddings function
                 def get_input_embeddings(self):