diff --git a/server/text_generation_server/models/vlm_causal_lm.py b/server/text_generation_server/models/vlm_causal_lm.py
index 06c9951e1..f17e9c247 100644
--- a/server/text_generation_server/models/vlm_causal_lm.py
+++ b/server/text_generation_server/models/vlm_causal_lm.py
@@ -784,6 +784,7 @@ class VlmCausalLM(Model):
             htorch.core.hpu_set_env()
 
         if world_size > 1:
+            os.environ.setdefault("DEEPSPEED_USE_HABANA_FRAMEWORKS_DETERMINISTIC_API", "1")
             model = self.get_deepspeed_model(
                 model_class, model_id, dtype, revision
             )