mmarch/text-generation-webui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Better separate the FlexGen case

Browse source
This commit is contained in:
oobabooga 2023-03-08 02:54:47 -03:00
parent 0e16c0bacb
commit 72d539dbff
1 changed files with 5 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

9
modules/text_generation.py
View file

@ -201,12 +201,8 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
reply = original_question + apply_extensions(reply[len(question):], "output") reply = original_question + apply_extensions(reply[len(question):], "output")
yield formatted_outputs(reply, shared.model_name) yield formatted_outputs(reply, shared.model_name)
if not shared.args.flexgen:
if output[-1] == n: if output[-1] == n:
break break
else:
if np.count_nonzero(input_ids[0] == n) < np.count_nonzero(output == n):
break
# Stream the output naively for FlexGen since it doesn't support 'stopping_criteria' # Stream the output naively for FlexGen since it doesn't support 'stopping_criteria'
else: else:
@ -223,11 +219,6 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
reply = original_question + apply_extensions(reply[len(question):], "output") reply = original_question + apply_extensions(reply[len(question):], "output")
yield formatted_outputs(reply, shared.model_name) yield formatted_outputs(reply, shared.model_name)
if not shared.args.flexgen:
if output[-1] == n:
break
input_ids = torch.reshape(output, (1, output.shape[0]))
else:
if np.count_nonzero(input_ids[0] == n) < np.count_nonzero(output == n): if np.count_nonzero(input_ids[0] == n) < np.count_nonzero(output == n):
break break
input_ids = np.reshape(output, (1, output.shape[0])) input_ids = np.reshape(output, (1, output.shape[0]))