Update raven_modeling_minimal.py

raven_modeling_minimal.py

@@ -242,6 +242,7 @@ class CausalSelfAttention(torch.nn.Module):
         if past_key_values is not None:
             k, v = past_key_values.update(k, v, step_idx)
 
+        return_attn = False # hard-code for now to rule out rushed eval errors
         if return_attn:
             y, attention_map = self.compute_eager_sdpa(q, k, v, attn_mask=mask)
         else: