fix: Update loss_algo attribute format in RLTestConfig (#1133)

RexBearIU · web-flow · commit c4fea25f0d9d · 2026-01-15T18:05:11.000+08:00
This pull request makes a small but important update to the RL training command generation logic, ensuring that the loss algorithm parameter is properly name-spaced for downstream compatibility.
diff --git a/dags/post_training/util/test_config_util.py b/dags/post_training/util/test_config_util.py
@@ -133,7 +133,7 @@ def generate_rl_training_command(
         f"tokenizer_path={self.tokenizer_path} "
         f"load_parameters_path={self.load_parameters_path} "
         f"base_output_directory={self.base_dir} "
-        f"loss_algo={loss_algo.loss_name}"
+        f"rl.loss_algo={loss_algo.loss_name}"
     )
 
     # Return as tuple for k8s yaml compatibility.

Original file line number	Diff line number	Diff line change
`@@ -133,7 +133,7 @@ def generate_rl_training_command(`
`133`	`133`	`f"tokenizer_path={self.tokenizer_path} "`
`134`	`134`	`f"load_parameters_path={self.load_parameters_path} "`
`135`	`135`	`f"base_output_directory={self.base_dir} "`
`136`		`- f"loss_algo={loss_algo.loss_name}"`
	`136`	`+ f"rl.loss_algo={loss_algo.loss_name}"`
`137`	`137`	`)`
`138`	`138`
`139`	`139`	`# Return as tuple for k8s yaml compatibility.`