diff --git a/deepspeed/inference/config.py b/deepspeed/inference/config.py
index 2ef43cb0239d..25994a110d31 100644
--- a/deepspeed/inference/config.py
+++ b/deepspeed/inference/config.py
@@ -310,6 +310,12 @@ def validate_dtype(cls, field_value, values):
             return field_value
         raise TypeError(f"Invalid type for dtype: {type(field_value)}")
 
+    @field_validator("max_out_tokens")
+    def validate_max_out_tokens(cls, field_value, values):
+        if field_value <= 0:
+            raise ValueError(f"max_out_tokens must be a positive integer, got {field_value}")
+        return field_value
+
     @field_validator("moe")
     def moe_backward_compat(cls, field_value, values):
         if isinstance(field_value, bool):
diff --git a/tests/unit/inference/test_inference_config.py b/tests/unit/inference/test_inference_config.py
index fcba7e0d601b..4e4a7d2ad664 100644
--- a/tests/unit/inference/test_inference_config.py
+++ b/tests/unit/inference/test_inference_config.py
@@ -82,3 +82,27 @@ def test_moe_backward_compat_bool(self):
             config = DeepSpeedInferenceConfig(moe=value)
             assert isinstance(config.moe, DeepSpeedMoEConfig)
             assert config.moe.enabled == value
+
+
+@pytest.mark.inference
+class TestInferenceConfigValidation:
+    """CPU-only validation tests for DeepSpeedInferenceConfig."""
+
+    def test_negative_max_out_tokens_rejected(self):
+        # Regression test for https://github.com/deepspeedai/DeepSpeed/issues/8339
+        from deepspeed.inference.config import DeepSpeedInferenceConfig
+
+        with pytest.raises(ValueError, match="max_out_tokens must be a positive integer"):
+            DeepSpeedInferenceConfig(max_out_tokens=-1)
+
+    def test_zero_max_out_tokens_rejected(self):
+        from deepspeed.inference.config import DeepSpeedInferenceConfig
+
+        with pytest.raises(ValueError, match="max_out_tokens must be a positive integer"):
+            DeepSpeedInferenceConfig(max_out_tokens=0)
+
+    def test_positive_max_out_tokens_accepted(self):
+        from deepspeed.inference.config import DeepSpeedInferenceConfig
+
+        config = DeepSpeedInferenceConfig(max_out_tokens=128)
+        assert config.max_out_tokens == 128