fix a off by one error with embedding at the start of the sentence

modules/sd_hijack.py

@@ -401,7 +401,7 @@ class EmbeddingsWithFixes(torch.nn.Module):
                 for offset, word in fixes:
                     emb = self.embeddings.word_embeddings[word]
                     emb_len = min(tensor.shape[0]-offset, emb.shape[0])
-                    tensor[offset:offset+emb_len] = self.embeddings.word_embeddings[word][0:emb_len]
+                    tensor[offset+1:offset+1+emb_len] = self.embeddings.word_embeddings[word][0:emb_len]
 
         return inputs_embeds