Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-09-20 18:45:09 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

transformers: Add eager attention option to make Gemma-2 work properly (#6188)

Browse source
This commit is contained in:
GralchemOz 2024-07-01 23:08:08 +08:00 committed by GitHub
parent 19a56dd538
commit 8a39f579d8
WARNING! Although there is a key with this ID in the database it does not verify this commit! This commit is SUSPICIOUS.
GPG key ID: B5690EEEBB952194
5 changed files with 7 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
modules/loaders.py
View file

@ -21,6 +21,7 @@ loaders_and_params = OrderedDict({
'trust_remote_code', 'trust_remote_code',
'no_use_fast', 'no_use_fast',
'use_flash_attention_2', 'use_flash_attention_2',
'use_eager_attention',
'alpha_value', 'alpha_value',
'compress_pos_emb', 'compress_pos_emb',
'disable_exllama', 'disable_exllama',

3
modules/models.py
View file

@ -146,6 +146,9 @@ def huggingface_loader(model_name):
if shared.args.force_safetensors: if shared.args.force_safetensors:
params['force_safetensors'] = True params['force_safetensors'] = True
if shared.args.use_eager_attention:
params['attn_implementation'] = 'eager'
config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=shared.args.trust_remote_code) config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=shared.args.trust_remote_code)
if 'chatglm' in model_name.lower(): if 'chatglm' in model_name.lower():

1
modules/shared.py
View file

@ -106,6 +106,7 @@ group.add_argument('--trust-remote-code', action='store_true', help='Set trust_r
group.add_argument('--force-safetensors', action='store_true', help='Set use_safetensors=True while loading the model. This prevents arbitrary code execution.') group.add_argument('--force-safetensors', action='store_true', help='Set use_safetensors=True while loading the model. This prevents arbitrary code execution.')
group.add_argument('--no_use_fast', action='store_true', help='Set use_fast=False while loading the tokenizer (it\'s True by default). Use this if you have any problems related to use_fast.') group.add_argument('--no_use_fast', action='store_true', help='Set use_fast=False while loading the tokenizer (it\'s True by default). Use this if you have any problems related to use_fast.')
group.add_argument('--use_flash_attention_2', action='store_true', help='Set use_flash_attention_2=True while loading the model.') group.add_argument('--use_flash_attention_2', action='store_true', help='Set use_flash_attention_2=True while loading the model.')
group.add_argument('--use_eager_attention', action='store_true', help='Set attn_implementation= eager while loading the model.')
# bitsandbytes 4-bit # bitsandbytes 4-bit
group = parser.add_argument_group('bitsandbytes 4-bit') group = parser.add_argument_group('bitsandbytes 4-bit')

1
modules/ui.py
View file

@ -69,6 +69,7 @@ def list_model_elements():
'trust_remote_code', 'trust_remote_code',
'no_use_fast', 'no_use_fast',
'use_flash_attention_2', 'use_flash_attention_2',
'use_eager_attention',
'load_in_4bit', 'load_in_4bit',
'compute_dtype', 'compute_dtype',
'quant_type', 'quant_type',

1
modules/ui_model_menu.py
View file

@ -115,6 +115,7 @@ def create_ui():
shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit", value=shared.args.load_in_4bit) shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit", value=shared.args.load_in_4bit)
shared.gradio['use_double_quant'] = gr.Checkbox(label="use_double_quant", value=shared.args.use_double_quant) shared.gradio['use_double_quant'] = gr.Checkbox(label="use_double_quant", value=shared.args.use_double_quant)
shared.gradio['use_flash_attention_2'] = gr.Checkbox(label="use_flash_attention_2", value=shared.args.use_flash_attention_2, info='Set use_flash_attention_2=True while loading the model.') shared.gradio['use_flash_attention_2'] = gr.Checkbox(label="use_flash_attention_2", value=shared.args.use_flash_attention_2, info='Set use_flash_attention_2=True while loading the model.')
shared.gradio['use_eager_attention'] = gr.Checkbox(label="use_eager_attention", value=shared.args.use_eager_attention, info='Set attn_implementation= eager while loading the model.')
shared.gradio['flash_attn'] = gr.Checkbox(label="flash_attn", value=shared.args.flash_attn, info='Use flash-attention.') shared.gradio['flash_attn'] = gr.Checkbox(label="flash_attn", value=shared.args.flash_attn, info='Use flash-attention.')
shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices) shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)
shared.gradio['tensorcores'] = gr.Checkbox(label="tensorcores", value=shared.args.tensorcores, info='NVIDIA only: use llama-cpp-python compiled with tensor cores support. This increases performance on RTX cards.') shared.gradio['tensorcores'] = gr.Checkbox(label="tensorcores", value=shared.args.tensorcores, info='NVIDIA only: use llama-cpp-python compiled with tensor cores support. This increases performance on RTX cards.')