Mirrors/text-generation-webui
1
0
Fork 0
mirror of https://github.com/oobabooga/text-generation-webui.git synced 2024-11-22 08:07:56 +01:00
Code Issues Packages Projects Releases Wiki Activity

Add --autosplit flag for ExLlamaV2 (#5524)

Browse Source
This commit is contained in:
oobabooga 2024-02-16 15:26:10 -03:00 committed by GitHub
parent 4039999be5
commit a6730f88f7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 29 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
modules/exllamav2.py
View File

@ -51,6 +51,14 @@ class Exllamav2Model:
model = ExLlamaV2(config)
if shared.args.cache_8bit:
cache = ExLlamaV2Cache_8bit(model, lazy=True)
else:
cache = ExLlamaV2Cache(model, lazy=True)
if shared.args.autosplit:
model.load_autosplit(cache)
else:
split = None
if shared.args.gpu_split:
split = [float(alloc) for alloc in shared.args.gpu_split.split(",")]
@ -58,11 +66,6 @@ class Exllamav2Model:
model.load(split)
tokenizer = ExLlamaV2Tokenizer(config)
if shared.args.cache_8bit:
cache = ExLlamaV2Cache_8bit(model)
else:
cache = ExLlamaV2Cache(model)
generator = ExLlamaV2StreamingGenerator(model, cache, tokenizer)
result = self()

18
modules/exllamav2_hf.py
View File

@ -37,18 +37,22 @@ class Exllamav2HF(PreTrainedModel):
super().__init__(PretrainedConfig())
self.ex_config = config
self.ex_model = ExLlamaV2(config)
self.loras = None
self.generation_config = GenerationConfig()
if shared.args.cache_8bit:
self.ex_cache = ExLlamaV2Cache_8bit(self.ex_model, lazy=True)
else:
self.ex_cache = ExLlamaV2Cache(self.ex_model, lazy=True)
if shared.args.autosplit:
self.ex_model.load_autosplit(self.ex_cache)
else:
split = None
if shared.args.gpu_split:
split = [float(alloc) for alloc in shared.args.gpu_split.split(",")]
self.ex_model.load(split)
self.generation_config = GenerationConfig()
self.loras = None
if shared.args.cache_8bit:
self.ex_cache = ExLlamaV2Cache_8bit(self.ex_model)
else:
self.ex_cache = ExLlamaV2Cache(self.ex_model)
self.past_seq = None
if shared.args.cfg_cache:

2
modules/loaders.py
View File

@ -78,6 +78,7 @@ loaders_and_params = OrderedDict({
'no_flash_attn',
'num_experts_per_token',
'cache_8bit',
'autosplit',
'alpha_value',
'compress_pos_emb',
'trust_remote_code',
@ -89,6 +90,7 @@ loaders_and_params = OrderedDict({
'no_flash_attn',
'num_experts_per_token',
'cache_8bit',
'autosplit',
'alpha_value',
'compress_pos_emb',
'exllamav2_info',

1
modules/shared.py
View File

@ -134,6 +134,7 @@ group.add_argument('--row_split', action='store_true', help='Split the model by
# ExLlamaV2
group = parser.add_argument_group('ExLlamaV2')
group.add_argument('--gpu-split', type=str, help='Comma-separated list of VRAM (in GB) to use per GPU device for model layers. Example: 20,7,7.')
group.add_argument('--autosplit', action='store_true', help='Autosplit the model tensors across the available GPUs. This causes --gpu-split to be ignored.')
group.add_argument('--max_seq_len', type=int, default=2048, help='Maximum sequence length.')
group.add_argument('--cfg-cache', action='store_true', help='ExLlamav2_HF: Create an additional cache for CFG negative prompts. Necessary to use CFG with that loader.')
group.add_argument('--no_flash_attn', action='store_true', help='Force flash-attention to not be used.')

1
modules/ui.py
View File

@ -76,6 +76,7 @@ def list_model_elements():
'no_flash_attn',
'num_experts_per_token',
'cache_8bit',
'autosplit',
'threads',
'threads_batch',
'n_batch',

1
modules/ui_model_menu.py
View File

@ -132,6 +132,7 @@ def create_ui():
shared.gradio['disk'] = gr.Checkbox(label="disk", value=shared.args.disk)
shared.gradio['bf16'] = gr.Checkbox(label="bf16", value=shared.args.bf16)
shared.gradio['cache_8bit'] = gr.Checkbox(label="cache_8bit", value=shared.args.cache_8bit, info='Use 8-bit cache to save VRAM.')
shared.gradio['autosplit'] = gr.Checkbox(label="autosplit", value=shared.args.autosplit, info='Automatically split the model tensors across the available GPUs.')
shared.gradio['no_flash_attn'] = gr.Checkbox(label="no_flash_attn", value=shared.args.no_flash_attn, info='Force flash-attention to not be used.')
shared.gradio['cfg_cache'] = gr.Checkbox(label="cfg-cache", value=shared.args.cfg_cache, info='Necessary to use CFG with this loader.')
shared.gradio['num_experts_per_token'] = gr.Number(label="Number of experts per token", value=shared.args.num_experts_per_token, info='Only applies to MoE models like Mixtral.')