Reject invalid ZenFlow top-k ratio endpoints

Signed-off-by: tandede <1090179959@qq.com>

Author: tandede

deepspeed/runtime/zenflow/zenflow_config.py

@@ -12,8 +12,8 @@
 class ZenFlowConfig(DeepSpeedConfigModel):
     """Configuration options for ZenFlow optimization module."""
 
-    topk_ratio: float = Field(0.1, ge=0.0, le=1.0)
-    """Ratio of top-k important gradient columns to retain (range: 0.0 to 1.0)."""
+    topk_ratio: float = Field(0.1, gt=0.0, lt=1.0)
+    """Ratio of top-k important gradient columns to retain (must be greater than 0.0 and less than 1.0)."""
 
     select_strategy: str = "auto"
     """Strategy for selecting important gradient indices.

tests/unit/runtime/zenflow/test_zf_config.py

@@ -59,6 +59,12 @@ def test_invalid_zenflow_type_raises():
         DeepSpeedZeroConfig(zenflow=123)
 
 
+@pytest.mark.parametrize("topk_ratio", [0.0, 1.0])
+def test_topk_ratio_excludes_endpoints(topk_ratio):
+    with pytest.raises(ValidationError):
+        ZenFlowConfig(topk_ratio=topk_ratio)
+
+
 def test_offload_and_zenflow_combined():
     """
     offload_optimizer and zenflow can be used together under stage 2