diff --git a/lora/lora.py b/lora/lora.py
index a90eda70..723e783d 100644
--- a/lora/lora.py
+++ b/lora/lora.py
@@ -220,8 +220,12 @@ def iterate_batches(dset, tokenizer, batch_size, train=False):
 def evaluate(model, dataset, loss, tokenizer, batch_size, num_batches):
     all_losses = []
     ntokens = 0
+
+    # num_batches can be -1 to indicate the entire set
+    index_iterator = iter(range(num_batches)) if num_batches != -1 else iter(int, 1)
+
     for it, batch in zip(
-        range(num_batches),
+        index_iterator,
         iterate_batches(dataset, tokenizer, batch_size),
     ):
         losses, toks = loss(model, *batch)