Validate positive inference min output tokens

min_out_tokens has no lower-bound validation, so a negative or zero value
passes config construction and is only caught later inside GenWorkSpace's
C++ workspace allocation. Add the same gt=0 constraint #8343 gave the
sibling max_out_tokens field.

Refs #8339

Signed-off-by: Amir Fathi <amirfathi.me@gmail.com>

Author: AmirF194

deepspeed/inference/config.py

@@ -265,7 +265,7 @@ class DeepSpeedInferenceConfig(DeepSpeedConfigModel):
     to the required token-length required for your use-case.
     """
 
-    min_out_tokens: int = Field(1, alias="min_tokens")
+    min_out_tokens: int = Field(1, alias="min_tokens", gt=0)
     """
     This argument communicates to the runtime the minimum number of tokens you
     expect you will need to generate. This will cause the runtime to error

tests/unit/inference/test_inference_config.py

@@ -6,10 +6,20 @@
 import pytest
 import torch
 import deepspeed
+from pydantic import ValidationError
+from deepspeed.inference.config import DeepSpeedInferenceConfig
 from unit.common import DistributedTest
 from unit.simple_model import create_config_from_dict
 
 
+@pytest.mark.inference
+@pytest.mark.parametrize("field", ["min_out_tokens", "min_tokens"])
+@pytest.mark.parametrize("value", [-1, 0])
+def test_min_out_tokens_must_be_positive(field, value):
+    with pytest.raises(ValidationError):
+        DeepSpeedInferenceConfig(**{field: value})
+
+
 @pytest.mark.inference
 class TestInferenceConfig(DistributedTest):
     world_size = 1