Merge pull request #3301 from AI-Hypercomputer:sanbao/format

Google-ML-Automation · Google-ML-Automation · commit 2670a9ecdc83 · 2026-03-06T11:13:32.000-08:00
PiperOrigin-RevId: 879723267
diff --git a/src/maxtext/trainers/post_train/rl/utils_rl.py b/src/maxtext/trainers/post_train/rl/utils_rl.py
@@ -105,7 +105,6 @@ def get_match_format_regex(tmvp_config):
   """Returns a compiled regex to extract the answer from a completion."""
   match_format = re.compile(
       (
-          r"^[\s]{0,}"
           rf"{tmvp_config.reasoning_start_token}.+{tmvp_config.reasoning_end_token}.*?"
           rf"{tmvp_config.solution_start_token}(.+?){tmvp_config.solution_end_token}"
       ),

Original file line number	Diff line number	Diff line change
`@@ -105,7 +105,6 @@ def get_match_format_regex(tmvp_config):`
`105`	`105`	`"""Returns a compiled regex to extract the answer from a completion."""`
`106`	`106`	`match_format = re.compile(`
`107`	`107`	`(`
`108`		`- r"^[\s]{0,}"`
`109`	`108`	`rf"{tmvp_config.reasoning_start_token}.+{tmvp_config.reasoning_end_token}.*?"`
`110`	`109`	`rf"{tmvp_config.solution_start_token}(.+?){tmvp_config.solution_end_token}"`
`111`	`110`	`),`