2023-06-17 00:00:37 +02:00
|
|
|
import functools
|
|
|
|
|
|
|
|
import gradio as gr
|
|
|
|
|
|
|
|
from modules import shared
|
|
|
|
|
|
|
|
loaders_and_params = {
|
|
|
|
'AutoGPTQ': [
|
|
|
|
'triton',
|
|
|
|
'no_inject_fused_attention',
|
|
|
|
'no_inject_fused_mlp',
|
2023-06-23 17:22:56 +02:00
|
|
|
'no_use_cuda_fp16',
|
2023-06-17 00:00:37 +02:00
|
|
|
'wbits',
|
|
|
|
'groupsize',
|
|
|
|
'desc_act',
|
|
|
|
'gpu_memory',
|
|
|
|
'cpu_memory',
|
|
|
|
'cpu',
|
|
|
|
'disk',
|
|
|
|
'auto_devices',
|
|
|
|
'trust_remote_code',
|
|
|
|
'autogptq_info',
|
|
|
|
],
|
|
|
|
'GPTQ-for-LLaMa': [
|
|
|
|
'wbits',
|
|
|
|
'groupsize',
|
|
|
|
'model_type',
|
|
|
|
'pre_layer',
|
|
|
|
'gptq_for_llama_info',
|
|
|
|
],
|
|
|
|
'llama.cpp': [
|
|
|
|
'n_ctx',
|
|
|
|
'n_gpu_layers',
|
|
|
|
'n_batch',
|
|
|
|
'threads',
|
|
|
|
'no_mmap',
|
2023-07-12 16:05:13 +02:00
|
|
|
'low_vram',
|
2023-06-17 00:00:37 +02:00
|
|
|
'mlock',
|
|
|
|
'llama_cpp_seed',
|
|
|
|
],
|
|
|
|
'Transformers': [
|
|
|
|
'cpu_memory',
|
|
|
|
'gpu_memory',
|
|
|
|
'trust_remote_code',
|
|
|
|
'load_in_8bit',
|
|
|
|
'bf16',
|
|
|
|
'cpu',
|
|
|
|
'disk',
|
|
|
|
'auto_devices',
|
|
|
|
'load_in_4bit',
|
|
|
|
'use_double_quant',
|
|
|
|
'quant_type',
|
|
|
|
'compute_dtype',
|
|
|
|
'trust_remote_code',
|
2023-06-17 00:07:16 +02:00
|
|
|
'transformers_info'
|
2023-06-17 00:00:37 +02:00
|
|
|
],
|
2023-07-12 20:33:25 +02:00
|
|
|
'ExLlama': [
|
2023-06-17 01:49:36 +02:00
|
|
|
'gpu_split',
|
2023-06-26 03:49:26 +02:00
|
|
|
'max_seq_len',
|
|
|
|
'compress_pos_emb',
|
2023-07-04 06:13:16 +02:00
|
|
|
'alpha_value',
|
2023-06-17 01:35:38 +02:00
|
|
|
'exllama_info',
|
2023-06-21 20:31:42 +02:00
|
|
|
],
|
2023-07-12 20:33:25 +02:00
|
|
|
'ExLlama_HF': [
|
2023-06-21 20:31:42 +02:00
|
|
|
'gpu_split',
|
2023-06-26 03:49:26 +02:00
|
|
|
'max_seq_len',
|
|
|
|
'compress_pos_emb',
|
2023-07-04 06:13:16 +02:00
|
|
|
'alpha_value',
|
2023-06-21 20:31:42 +02:00
|
|
|
'exllama_HF_info',
|
2023-06-17 01:35:38 +02:00
|
|
|
]
|
2023-06-17 00:00:37 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
def get_gpu_memory_keys():
|
|
|
|
return [k for k in shared.gradio if k.startswith('gpu_memory')]
|
|
|
|
|
|
|
|
|
|
|
|
@functools.cache
|
|
|
|
def get_all_params():
|
|
|
|
all_params = set()
|
|
|
|
for k in loaders_and_params:
|
|
|
|
for el in loaders_and_params[k]:
|
|
|
|
all_params.add(el)
|
|
|
|
|
|
|
|
if 'gpu_memory' in all_params:
|
|
|
|
all_params.remove('gpu_memory')
|
|
|
|
for k in get_gpu_memory_keys():
|
|
|
|
all_params.add(k)
|
|
|
|
|
|
|
|
return sorted(all_params)
|
|
|
|
|
|
|
|
|
|
|
|
def make_loader_params_visible(loader):
|
|
|
|
params = []
|
|
|
|
all_params = get_all_params()
|
|
|
|
if loader in loaders_and_params:
|
|
|
|
params = loaders_and_params[loader]
|
|
|
|
|
|
|
|
if 'gpu_memory' in params:
|
|
|
|
params.remove('gpu_memory')
|
|
|
|
params += get_gpu_memory_keys()
|
|
|
|
|
|
|
|
return [gr.update(visible=True) if k in params else gr.update(visible=False) for k in all_params]
|