Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-26 01:30:20 +01:00
Code Issues Packages Projects Releases 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
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
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:
@ -223,14 +219,9 @@ 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 np.count_nonzero(input_ids[0] == n) < np.count_nonzero(output == n):
if output[-1] == n: break
break input_ids = np.reshape(output, (1, output.shape[0]))
input_ids = torch.reshape(output, (1, output.shape[0]))
else:
if np.count_nonzero(input_ids[0] == n) < np.count_nonzero(output == n):
break
input_ids = np.reshape(output, (1, output.shape[0]))
if shared.soft_prompt: if shared.soft_prompt:
inputs_embeds, filler_input_ids = generate_softprompt_input_tensors(input_ids) inputs_embeds, filler_input_ids = generate_softprompt_input_tensors(input_ids)