fix hardcode

Author: LovelyBuggies

configs/grpo_che_config.yaml

@@ -37,6 +37,8 @@ grpo:
   learning_rate: 2.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   joint_mode: aligned

configs/grpo_he_config.yaml

@@ -37,6 +37,8 @@ grpo:
   learning_rate: 2.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   joint_mode: aligned

configs/grpo_mbpp_config.yaml

@@ -37,6 +37,8 @@ grpo:
   learning_rate: 3.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   joint_mode: aligned

configs/magrpo_che_config.yaml

@@ -37,6 +37,8 @@ magrpo:
   learning_rate: 2.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   temperature: 0.8

configs/magrpo_he_config.yaml

@@ -37,6 +37,8 @@ magrpo:
   learning_rate: 2.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   joint_mode: aligned

configs/magrpo_mbpp_config.yaml

@@ -37,6 +37,8 @@ magrpo:
   learning_rate: 3.0e-5
   logging_steps: 50
   save_steps: 200
+  eval_interval: 4
+  eval_num_samples: 4
   num_generations: 4
   max_new_tokens: 256
   temperature: 0.8

train_grpo.py

@@ -357,6 +357,8 @@ def _resolver(prompt: str):
         learning_rate=grpo_config.get("learning_rate", 2e-5),
         logging_steps=grpo_config.get("logging_steps", 50),
         save_steps=grpo_config.get("save_steps", 200),
+        eval_interval=grpo_config.get("eval_interval", 4),
+        eval_num_samples=grpo_config.get("eval_num_samples", 4),
         num_generations=grpo_config.get("num_generations", 4),
         max_new_tokens=grpo_config.get("max_new_tokens", 256),
         temperature=temperature,

train_magrpo.py

@@ -410,6 +410,8 @@ def _resolver(prompt: str):
         learning_rate=magrpo_config.get("learning_rate", 2e-5),
         logging_steps=magrpo_config.get("logging_steps", 50),
         save_steps=magrpo_config.get("save_steps", 200),
+        eval_interval=magrpo_config.get("eval_interval", 4),
+        eval_num_samples=magrpo_config.get("eval_num_samples", 4),
         num_generations=magrpo_config.get("num_generations", 4),
         max_new_tokens=magrpo_config.get("max_new_tokens", 256),
         temperature=temperature,