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

Add alpha_value/compress_pos_emb to ExLlama-v2

Browse source
This commit is contained in:
oobabooga 2023-09-12 15:02:47 -07:00
parent 90fca6a77d
commit 18e6b275f3
3 changed files with 10 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

3
modules/exllamav2.py
View file

@ -28,6 +28,9 @@ class Exllamav2Model:
config.prepare()
config.max_seq_len = shared.args.max_seq_len
config.rope_scale = shared.args.compress_pos_emb
config.rope_alpha = shared.args.alpha_value
model = ExLlamaV2(config)
split = None

3
modules/exllamav2_hf.py
View file

@ -114,6 +114,9 @@ class Exllamav2HF(PreTrainedModel):
config = ExLlamaV2Config()
config.model_dir = pretrained_model_name_or_path
config.prepare()
config.max_seq_len = shared.args.max_seq_len
config.rope_scale = shared.args.compress_pos_emb
config.rope_alpha = shared.args.alpha_value
return Exllamav2HF(config)

4
modules/loaders.py
View file

@ -45,11 +45,15 @@ loaders_and_params = OrderedDict({
'ExLlamav2': [
'gpu_split',
'max_seq_len',
'alpha_value',
'compress_pos_emb',
],
'ExLlamav2_HF': [
'gpu_split',
'max_seq_len',
'cfg_cache',
'alpha_value',
'compress_pos_emb',
],
'AutoGPTQ': [
'triton',