diff --git a/examples/hstu/configs/inference_config.py b/examples/hstu/configs/inference_config.py
index 6f44362c6..a89850b19 100755
--- a/examples/hstu/configs/inference_config.py
+++ b/examples/hstu/configs/inference_config.py
@@ -262,16 +262,11 @@ def get_kvcache_metadata_buffer(
     hstu_config: InferenceHSTUConfig, kvcache_config: KVCacheConfig
 ):
     device = torch.cuda.current_device()
-    torch.bfloat16 if hstu_config.bf16 else torch.float16 if hstu_config.fp16 else torch.float32
 
     max_new_history_seqlen = hstu_config.max_batch_size * hstu_config.max_seq_len
     max_num_pages_per_seq = (
         hstu_config.max_seq_len + hstu_config.max_seq_len + kvcache_config.page_size - 1
     ) // kvcache_config.page_size
-    max_host_kv_buffer_size = (
-        hstu_config.max_batch_size * hstu_config.max_seq_len,
-        hstu_config.num_heads * hstu_config.head_dim,
-    )
 
     default_num_pages_per_seq = 4
     paged_indices_buffer = torch.randint(