Fix bug in multigpu setups

This commit is contained in:
oobabooga 2023-03-06 14:58:30 -03:00
parent 5bed607b77
commit d88b7836c6

View File

@ -194,7 +194,7 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
yield formatted_outputs(reply, shared.model_name)
if not shared.args.flexgen:
if output[-1] == n:
if int(output[-1]) == int(n):
break
input_ids = torch.reshape(output, (1, output.shape[0]))
else: