diff --git a/nanochat/gpt.py b/nanochat/gpt.py
index 208acd1..5487047 100644
--- a/nanochat/gpt.py
+++ b/nanochat/gpt.py
@@ -416,7 +416,7 @@ class GPT(nn.Module):
         if targets is not None:
             # training: given the targets, compute and return the loss
             # TODO experiment with chunked cross-entropy?
-            loss = F.cross_entropy(logits.view(-1, logits.size(-1)), targets.view(-1), ignore_index=-1, reduction=loss_reduction)
+            loss = F.cross_entropy(logits.reshape(-1, logits.size(-1)), targets.reshape(-1), ignore_index=-1, reduction=loss_reduction)
             return loss
         else:
             # inference: just return the logits directly
diff --git a/nanochat/loss_eval.py b/nanochat/loss_eval.py
index 5a556e6..6f59b11 100644
--- a/nanochat/loss_eval.py
+++ b/nanochat/loss_eval.py
@@ -32,7 +32,7 @@ def evaluate_bpb(model, batches, steps, token_bytes):
         x, y = next(batch_iter)
         loss2d = model(x, y, loss_reduction='none') # (B, T)
         loss2d = loss2d.view(-1) # flatten
-        y = y.view(-1) # flatten
+        y = y.reshape(-1) # flatten
         if (y.int() < 0).any(): # mps does not currently have kernel for < 0 for int64, only int32
             # slightly more complex code path if some target tokens are ignore_index (e.g. -1)
             # any target token < 0 is to be ignored: do NOT index token_bytes with negatives