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

Disable logits_all in llamacpp_HF (makes processing 3x faster)

Browse source
This commit is contained in:
oobabooga 2023-11-07 14:35:48 -08:00
parent 5c3eb22ce6
commit af3d25a503
5 changed files with 5 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
modules/llamacpp_hf.py
View file

@ -204,7 +204,7 @@ class LlamacppHF(PreTrainedModel):
'rope_freq_base': RoPE.get_rope_freq_base(shared.args.alpha_value, shared.args.rope_freq_base),
'tensor_split': tensor_split_list,
'rope_freq_scale': 1.0 / shared.args.compress_pos_emb,
'logits_all': True,
'logits_all': shared.args.logits_all,
}
Llama = llama_cpp_lib().Llama