Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-25 17:29:22 +01:00
Code Issues Packages Projects Releases Wiki Activity

Fix bug in multigpu setups (attempt #2)

Browse Source
This commit is contained in:
oobabooga 2023-03-06 15:23:29 -03:00
parent d88b7836c6
commit 24c4c20391
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
modules/text_generation.py
View File

@ -112,7 +112,7 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
input_ids = encode(question, max_new_tokens) input_ids = encode(question, max_new_tokens)
cuda = "" if (shared.args.cpu or shared.args.deepspeed or shared.args.flexgen) else ".cuda()" cuda = "" if (shared.args.cpu or shared.args.deepspeed or shared.args.flexgen) else ".cuda()"
n = shared.tokenizer.eos_token_id if eos_token is None else encode(eos_token)[0][-1] n = torch.tensor(shared.tokenizer.eos_token_id) if eos_token is None else encode(eos_token)[0][-1]
if stopping_string is not None: if stopping_string is not None:
# The stopping_criteria code below was copied from # The stopping_criteria code below was copied from
# https://github.com/PygmalionAI/gradio-ui/blob/master/src/model.py # https://github.com/PygmalionAI/gradio-ui/blob/master/src/model.py
@ -194,7 +194,7 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
yield formatted_outputs(reply, shared.model_name) yield formatted_outputs(reply, shared.model_name)
if not shared.args.flexgen: if not shared.args.flexgen:
if int(output[-1]) == int(n): if output[-1].to("cpu") == n.to("cpu"):
break break
input_ids = torch.reshape(output, (1, output.shape[0])) input_ids = torch.reshape(output, (1, output.shape[0]))
else: else: