Fix Wrong Text-encoder Grad Setting in Custom_Diffusion Training (#7302)

fix index in set textencoder grad
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

examples/custom_diffusion/train_custom_diffusion.py

@@ -1178,7 +1178,7 @@ def main(args):
                         grads_text_encoder = text_encoder.get_input_embeddings().weight.grad
                     # Get the index for tokens that we want to zero the grads for
                     index_grads_to_zero = torch.arange(len(tokenizer)) != modifier_token_id[0]
-                    for i in range(len(modifier_token_id[1:])):
+                    for i in range(1, len(modifier_token_id)):
                         index_grads_to_zero = index_grads_to_zero & (
                             torch.arange(len(tokenizer)) != modifier_token_id[i]
                         )