diff --git a/unsloth/models/rl.py b/unsloth/models/rl.py
index 7b363d8fc1..d53c9606d2 100644
--- a/unsloth/models/rl.py
+++ b/unsloth/models/rl.py
@@ -112,12 +112,12 @@ class Unsloth{RLConfig_name}({RLConfig_name}):
     """
     {__RLConfig_doc__}
     """
-    sampling_params: Optional[Any] = field(
+    vllm_sampling_params: Optional[Any] = field(
         default = None,
         metadata = {{'help': 'vLLM SamplingParams'}},
     )
     def __init__({RLConfig_arguments},
-        sampling_params = None,
+        vllm_sampling_params = None,
         **kwargs,
     ):
 {RLConfig_extra_args}
diff --git a/unsloth/models/rl_replacements.py b/unsloth/models/rl_replacements.py
index b2501c94fc..b9ba34726a 100644
--- a/unsloth/models/rl_replacements.py
+++ b/unsloth/models/rl_replacements.py
@@ -188,8 +188,8 @@ def grpo_trainer__get_per_token_logps(function_name, function):
             # For transformers<=4.48, logits_to_keep argument isn't supported, so here we drop logits ourselves.
             # See https://github.com/huggingface/trl/issues/2770
             logits = logits[:, -logits_to_keep:]
-            return logits
-            # return selective_log_softmax(logits, input_ids)  #  compute logprobs for the input tokens
+            # return logits
+            return selective_log_softmax(logits, input_ids)  #  compute logprobs for the input tokens
         pass
     pass
 
@@ -199,7 +199,7 @@ pass
 RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer__get_per_token_logps)
 
 grpo_compute_loss = RL_REPLACEMENTS["grpo_compute_loss"]
-RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(grpo_compute_loss))
+# RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(grpo_compute_loss))
 
 # Edit _get_per_token_logps to handle mixed precision
 def grpo_trainer_compute_loss(function_name, function):
@@ -245,7 +245,7 @@ def grpo_trainer_compute_loss(function_name, function):
     function = inspect.getsource(compute_loss)
     return function
 pass
-RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer_compute_loss)
+# RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer_compute_loss)
 
 # https://github.com/huggingface/trl/blob/main/trl/trainer/grpo_trainer.py#L356
 # TRL warns if batch size is not a multiple of num_generations -> fix this.