diff --git a/vllm/model_executor/layers/sampler.py b/vllm/model_executor/layers/sampler.py
index f9d95fa7..25ba48c2 100644
--- a/vllm/model_executor/layers/sampler.py
+++ b/vllm/model_executor/layers/sampler.py
@@ -112,27 +112,6 @@ def _prune_hidden_states(
                                       sampling_metadata.selected_token_indices)
 
 
-def _get_prompt_and_output_tokens(
-    sampling_metadata: SamplingMetadata,
-) -> Tuple[List[List[int]], List[List[int]]]:
-    prompt_tokens: List[List[int]] = []
-    output_tokens: List[List[int]] = []
-    for i, seq_group in enumerate(sampling_metadata.seq_groups):
-        seq_ids, sampling_params = seq_group
-        if (i < sampling_metadata.num_prompts
-                and sampling_params.prompt_logprobs is not None):
-            # NOTE: prompt token positions do not need output tokens to
-            # compute penalties.
-            prompt_len = sampling_metadata.prompt_lens[i]
-            prompt_tokens.extend([] for _ in range(prompt_len - 1))
-            output_tokens.extend([] for _ in range(prompt_len - 1))
-        for seq_id in seq_ids:
-            seq_data = sampling_metadata.seq_data[seq_id]
-            prompt_tokens.append(seq_data.prompt_token_ids)
-            output_tokens.append(seq_data.output_token_ids)
-    return prompt_tokens, output_tokens
-
-
 def _get_bin_counts_and_mask(
     tokens: torch.Tensor,
     vocab_size: int,
diff --git a/vllm/sampling_params.py b/vllm/sampling_params.py
index 30a8036a..b5710eef 100644
--- a/vllm/sampling_params.py
+++ b/vllm/sampling_params.py
@@ -100,7 +100,7 @@ class SamplingParams:
         temperature: float = 1.0,
         top_p: float = 1.0,
         top_k: int = -1,
-        min_p: int = 0.0,
+        min_p: float = 0.0,
         use_beam_search: bool = False,
         length_penalty: float = 1.0,
         early_stopping: Union[bool, str] = False,