Spaces:
Runtime error
Runtime error
Upload 94 files
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- modules/AutoGPTQ_loader.py +74 -0
- modules/GPTQ_loader.py +171 -0
- modules/LoRA.py +153 -0
- modules/RoPE.py +18 -0
- modules/__pycache__/LoRA.cpython-311.pyc +0 -0
- modules/__pycache__/RoPE.cpython-311.pyc +0 -0
- modules/__pycache__/block_requests.cpython-311.pyc +0 -0
- modules/__pycache__/callbacks.cpython-311.pyc +0 -0
- modules/__pycache__/chat.cpython-311.pyc +0 -0
- modules/__pycache__/evaluate.cpython-311.pyc +0 -0
- modules/__pycache__/extensions.cpython-311.pyc +0 -0
- modules/__pycache__/github.cpython-311.pyc +0 -0
- modules/__pycache__/html_generator.cpython-311.pyc +0 -0
- modules/__pycache__/llamacpp_hf.cpython-311.pyc +0 -0
- modules/__pycache__/llamacpp_model.cpython-311.pyc +0 -0
- modules/__pycache__/loaders.cpython-311.pyc +0 -0
- modules/__pycache__/logging_colors.cpython-311.pyc +0 -0
- modules/__pycache__/logits.cpython-311.pyc +0 -0
- modules/__pycache__/metadata_gguf.cpython-311.pyc +0 -0
- modules/__pycache__/models.cpython-311.pyc +0 -0
- modules/__pycache__/models_settings.cpython-311.pyc +0 -0
- modules/__pycache__/one_click_installer_check.cpython-311.pyc +0 -0
- modules/__pycache__/presets.cpython-311.pyc +0 -0
- modules/__pycache__/prompts.cpython-311.pyc +0 -0
- modules/__pycache__/relative_imports.cpython-311.pyc +0 -0
- modules/__pycache__/sampler_hijack.cpython-311.pyc +0 -0
- modules/__pycache__/shared.cpython-311.pyc +0 -0
- modules/__pycache__/text_generation.cpython-311.pyc +0 -0
- modules/__pycache__/training.cpython-311.pyc +0 -0
- modules/__pycache__/ui.cpython-311.pyc +0 -0
- modules/__pycache__/ui_chat.cpython-311.pyc +0 -0
- modules/__pycache__/ui_default.cpython-311.pyc +0 -0
- modules/__pycache__/ui_file_saving.cpython-311.pyc +0 -0
- modules/__pycache__/ui_model_menu.cpython-311.pyc +0 -0
- modules/__pycache__/ui_notebook.cpython-311.pyc +0 -0
- modules/__pycache__/ui_parameters.cpython-311.pyc +0 -0
- modules/__pycache__/ui_session.cpython-311.pyc +0 -0
- modules/__pycache__/utils.cpython-311.pyc +0 -0
- modules/block_requests.py +47 -0
- modules/callbacks.py +103 -0
- modules/chat.py +927 -0
- modules/ctransformers_model.py +79 -0
- modules/deepspeed_parameters.py +74 -0
- modules/evaluate.py +153 -0
- modules/exllamav2.py +149 -0
- modules/exllamav2_hf.py +170 -0
- modules/extensions.py +232 -0
- modules/github.py +38 -0
- modules/grammar/__pycache__/grammar_utils.cpython-311.pyc +0 -0
- modules/grammar/__pycache__/logits_process.cpython-311.pyc +0 -0
modules/AutoGPTQ_loader.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from pathlib import Path
|
2 |
+
|
3 |
+
from accelerate.utils import is_xpu_available
|
4 |
+
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
|
5 |
+
|
6 |
+
import modules.shared as shared
|
7 |
+
from modules.logging_colors import logger
|
8 |
+
from modules.models import get_max_memory_dict
|
9 |
+
|
10 |
+
|
11 |
+
def load_quantized(model_name):
|
12 |
+
path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
|
13 |
+
pt_path = None
|
14 |
+
|
15 |
+
# Find the model checkpoint
|
16 |
+
if shared.args.checkpoint:
|
17 |
+
pt_path = Path(shared.args.checkpoint)
|
18 |
+
else:
|
19 |
+
for ext in ['.safetensors', '.pt', '.bin']:
|
20 |
+
found = list(path_to_model.glob(f"*{ext}"))
|
21 |
+
if len(found) > 0:
|
22 |
+
if len(found) > 1:
|
23 |
+
logger.warning(f'More than one {ext} model has been found. The last one will be selected. It could be wrong.')
|
24 |
+
|
25 |
+
pt_path = found[-1]
|
26 |
+
break
|
27 |
+
|
28 |
+
if pt_path is None:
|
29 |
+
logger.error("The model could not be loaded because its checkpoint file in .bin/.pt/.safetensors format could not be located.")
|
30 |
+
return
|
31 |
+
|
32 |
+
use_safetensors = pt_path.suffix == '.safetensors'
|
33 |
+
if not (path_to_model / "quantize_config.json").exists():
|
34 |
+
quantize_config = BaseQuantizeConfig(
|
35 |
+
bits=bits if (bits := shared.args.wbits) > 0 else 4,
|
36 |
+
group_size=gs if (gs := shared.args.groupsize) > 0 else -1,
|
37 |
+
desc_act=shared.args.desc_act
|
38 |
+
)
|
39 |
+
else:
|
40 |
+
quantize_config = None
|
41 |
+
|
42 |
+
# Define the params for AutoGPTQForCausalLM.from_quantized
|
43 |
+
params = {
|
44 |
+
'model_basename': pt_path.stem,
|
45 |
+
'device': "xpu:0" if is_xpu_available() else "cuda:0" if not shared.args.cpu else "cpu",
|
46 |
+
'use_triton': shared.args.triton,
|
47 |
+
'inject_fused_attention': not shared.args.no_inject_fused_attention,
|
48 |
+
'inject_fused_mlp': not shared.args.no_inject_fused_mlp,
|
49 |
+
'use_safetensors': use_safetensors,
|
50 |
+
'trust_remote_code': shared.args.trust_remote_code,
|
51 |
+
'max_memory': get_max_memory_dict(),
|
52 |
+
'quantize_config': quantize_config,
|
53 |
+
'use_cuda_fp16': not shared.args.no_use_cuda_fp16,
|
54 |
+
'disable_exllama': shared.args.disable_exllama,
|
55 |
+
'disable_exllamav2': shared.args.disable_exllamav2,
|
56 |
+
}
|
57 |
+
|
58 |
+
logger.info(f"The AutoGPTQ params are: {params}")
|
59 |
+
model = AutoGPTQForCausalLM.from_quantized(path_to_model, **params)
|
60 |
+
|
61 |
+
# These lines fix the multimodal extension when used with AutoGPTQ
|
62 |
+
if hasattr(model, 'model'):
|
63 |
+
if not hasattr(model, 'dtype'):
|
64 |
+
if hasattr(model.model, 'dtype'):
|
65 |
+
model.dtype = model.model.dtype
|
66 |
+
|
67 |
+
if hasattr(model.model, 'model') and hasattr(model.model.model, 'embed_tokens'):
|
68 |
+
if not hasattr(model, 'embed_tokens'):
|
69 |
+
model.embed_tokens = model.model.model.embed_tokens
|
70 |
+
|
71 |
+
if not hasattr(model.model, 'embed_tokens'):
|
72 |
+
model.model.embed_tokens = model.model.model.embed_tokens
|
73 |
+
|
74 |
+
return model
|
modules/GPTQ_loader.py
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import inspect
|
2 |
+
import re
|
3 |
+
from pathlib import Path
|
4 |
+
|
5 |
+
import accelerate
|
6 |
+
import torch
|
7 |
+
import transformers
|
8 |
+
from accelerate.utils import is_xpu_available
|
9 |
+
from gptq_for_llama import llama_inference_offload
|
10 |
+
from gptq_for_llama.modelutils import find_layers
|
11 |
+
from gptq_for_llama.quant import make_quant
|
12 |
+
from transformers import AutoConfig, AutoModelForCausalLM
|
13 |
+
|
14 |
+
import modules.shared as shared
|
15 |
+
from modules.logging_colors import logger
|
16 |
+
|
17 |
+
|
18 |
+
# This function is a replacement for the load_quant function in the
|
19 |
+
# GPTQ-for_LLaMa repository. It supports more models and branches.
|
20 |
+
def _load_quant(model, checkpoint, wbits, groupsize=-1, faster_kernel=False, exclude_layers=None, kernel_switch_threshold=128, eval=True):
|
21 |
+
exclude_layers = exclude_layers or ['lm_head']
|
22 |
+
|
23 |
+
def noop(*args, **kwargs):
|
24 |
+
pass
|
25 |
+
|
26 |
+
config = AutoConfig.from_pretrained(model, trust_remote_code=shared.args.trust_remote_code)
|
27 |
+
torch.nn.init.kaiming_uniform_ = noop
|
28 |
+
torch.nn.init.uniform_ = noop
|
29 |
+
torch.nn.init.normal_ = noop
|
30 |
+
|
31 |
+
torch.set_default_dtype(torch.half)
|
32 |
+
transformers.modeling_utils._init_weights = False
|
33 |
+
torch.set_default_dtype(torch.half)
|
34 |
+
model = AutoModelForCausalLM.from_config(config, trust_remote_code=shared.args.trust_remote_code)
|
35 |
+
torch.set_default_dtype(torch.float)
|
36 |
+
if eval:
|
37 |
+
model = model.eval()
|
38 |
+
|
39 |
+
layers = find_layers(model)
|
40 |
+
for name in exclude_layers:
|
41 |
+
if name in layers:
|
42 |
+
del layers[name]
|
43 |
+
|
44 |
+
gptq_args = inspect.getfullargspec(make_quant).args
|
45 |
+
|
46 |
+
make_quant_kwargs = {
|
47 |
+
'module': model,
|
48 |
+
'names': layers,
|
49 |
+
'bits': wbits,
|
50 |
+
}
|
51 |
+
if 'groupsize' in gptq_args:
|
52 |
+
make_quant_kwargs['groupsize'] = groupsize
|
53 |
+
if 'faster' in gptq_args:
|
54 |
+
make_quant_kwargs['faster'] = faster_kernel
|
55 |
+
if 'kernel_switch_threshold' in gptq_args:
|
56 |
+
make_quant_kwargs['kernel_switch_threshold'] = kernel_switch_threshold
|
57 |
+
|
58 |
+
make_quant(**make_quant_kwargs)
|
59 |
+
|
60 |
+
del layers
|
61 |
+
if checkpoint.endswith('.safetensors'):
|
62 |
+
from safetensors.torch import load_file as safe_load
|
63 |
+
model.load_state_dict(safe_load(checkpoint), strict=False)
|
64 |
+
else:
|
65 |
+
model.load_state_dict(torch.load(checkpoint, weights_only=True), strict=False)
|
66 |
+
|
67 |
+
model.seqlen = 2048
|
68 |
+
return model
|
69 |
+
|
70 |
+
|
71 |
+
# Used to locate the .pt/.safetensors quantized file
|
72 |
+
def find_quantized_model_file(model_name):
|
73 |
+
if shared.args.checkpoint:
|
74 |
+
return Path(shared.args.checkpoint)
|
75 |
+
|
76 |
+
path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
|
77 |
+
pt_path = None
|
78 |
+
priority_name_list = [
|
79 |
+
Path(f'{shared.args.model_dir}/{model_name}{hyphen}{shared.args.wbits}bit{group}{ext}')
|
80 |
+
for group in ([f'-{shared.args.groupsize}g', ''] if shared.args.groupsize > 0 else [''])
|
81 |
+
for ext in ['.safetensors', '.pt']
|
82 |
+
for hyphen in ['-', f'/{model_name}-', '/']
|
83 |
+
]
|
84 |
+
|
85 |
+
for path in priority_name_list:
|
86 |
+
if path.exists():
|
87 |
+
pt_path = path
|
88 |
+
break
|
89 |
+
|
90 |
+
# If the model hasn't been found with a well-behaved name, pick the last .pt
|
91 |
+
# or the last .safetensors found in its folder as a last resort
|
92 |
+
if not pt_path:
|
93 |
+
for ext in ['.pt', '.safetensors']:
|
94 |
+
found = list(path_to_model.glob(f"*{ext}"))
|
95 |
+
if len(found) > 0:
|
96 |
+
if len(found) > 1:
|
97 |
+
logger.warning(f'More than one {ext} model has been found. The last one will be selected. It could be wrong.')
|
98 |
+
|
99 |
+
pt_path = found[-1]
|
100 |
+
break
|
101 |
+
|
102 |
+
return pt_path
|
103 |
+
|
104 |
+
|
105 |
+
# The function that loads the model in modules/models.py
|
106 |
+
def load_quantized(model_name):
|
107 |
+
if shared.args.model_type is None:
|
108 |
+
logger.error("The model could not be loaded because its type could not be inferred from its name.")
|
109 |
+
logger.error("Please specify the type manually using the --model_type argument.")
|
110 |
+
return None
|
111 |
+
|
112 |
+
# Select the appropriate load_quant function
|
113 |
+
model_type = shared.args.model_type.lower()
|
114 |
+
if shared.args.pre_layer and model_type == 'llama':
|
115 |
+
load_quant = llama_inference_offload.load_quant
|
116 |
+
elif model_type in ('llama', 'opt', 'gptj'):
|
117 |
+
if shared.args.pre_layer:
|
118 |
+
logger.warning("Ignoring --pre_layer because it only works for llama model type.")
|
119 |
+
|
120 |
+
load_quant = _load_quant
|
121 |
+
else:
|
122 |
+
logger.error("Unknown pre-quantized model type specified. Only 'llama', 'opt' and 'gptj' are supported")
|
123 |
+
exit()
|
124 |
+
|
125 |
+
# Find the quantized model weights file (.pt/.safetensors)
|
126 |
+
path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
|
127 |
+
pt_path = find_quantized_model_file(model_name)
|
128 |
+
if not pt_path:
|
129 |
+
logger.error("Could not find the quantized model in .pt or .safetensors format. Exiting.")
|
130 |
+
exit()
|
131 |
+
else:
|
132 |
+
logger.info(f"Found the following quantized model: {pt_path}")
|
133 |
+
|
134 |
+
# qwopqwop200's offload
|
135 |
+
if model_type == 'llama' and shared.args.pre_layer:
|
136 |
+
if len(shared.args.pre_layer) == 1:
|
137 |
+
pre_layer = shared.args.pre_layer[0]
|
138 |
+
else:
|
139 |
+
pre_layer = shared.args.pre_layer
|
140 |
+
|
141 |
+
model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits, shared.args.groupsize, pre_layer)
|
142 |
+
else:
|
143 |
+
threshold = False if model_type == 'gptj' else 128
|
144 |
+
model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits, shared.args.groupsize, kernel_switch_threshold=threshold)
|
145 |
+
|
146 |
+
# accelerate offload (doesn't work properly)
|
147 |
+
if shared.args.gpu_memory or torch.cuda.device_count() > 1 or (is_xpu_available() and torch.xpu.device_count() > 1):
|
148 |
+
if shared.args.gpu_memory:
|
149 |
+
memory_map = list(map(lambda x: x.strip(), shared.args.gpu_memory))
|
150 |
+
max_cpu_memory = shared.args.cpu_memory.strip() if shared.args.cpu_memory is not None else '99GiB'
|
151 |
+
max_memory = {}
|
152 |
+
for i in range(len(memory_map)):
|
153 |
+
max_memory[i] = f'{memory_map[i]}GiB' if not re.match('.*ib$', memory_map[i].lower()) else memory_map[i]
|
154 |
+
|
155 |
+
max_memory['cpu'] = f'{max_cpu_memory}GiB' if not re.match('.*ib$', max_cpu_memory.lower()) else max_cpu_memory
|
156 |
+
else:
|
157 |
+
max_memory = accelerate.utils.get_balanced_memory(model)
|
158 |
+
|
159 |
+
device_map = accelerate.infer_auto_device_map(model, max_memory=max_memory, no_split_module_classes=["LlamaDecoderLayer"])
|
160 |
+
logger.info("Using the following device map for the quantized model:", device_map)
|
161 |
+
# https://huggingface.co/docs/accelerate/package_reference/big_modeling#accelerate.dispatch_model
|
162 |
+
model = accelerate.dispatch_model(model, device_map=device_map, offload_buffers=True)
|
163 |
+
|
164 |
+
# No offload
|
165 |
+
elif not shared.args.cpu:
|
166 |
+
if is_xpu_available():
|
167 |
+
model = model.to(torch.device("xpu:0"))
|
168 |
+
else:
|
169 |
+
model = model.to(torch.device('cuda:0'))
|
170 |
+
|
171 |
+
return model
|
modules/LoRA.py
ADDED
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from pathlib import Path
|
2 |
+
|
3 |
+
import torch
|
4 |
+
from peft import PeftModel
|
5 |
+
from transformers import is_torch_xpu_available
|
6 |
+
|
7 |
+
import modules.shared as shared
|
8 |
+
from modules.logging_colors import logger
|
9 |
+
from modules.models import reload_model
|
10 |
+
|
11 |
+
|
12 |
+
def add_lora_to_model(lora_names):
|
13 |
+
if 'GPTQForCausalLM' in shared.model.__class__.__name__ or shared.args.loader == 'AutoGPTQ':
|
14 |
+
add_lora_autogptq(lora_names)
|
15 |
+
elif shared.model.__class__.__name__ in ['Exllamav2Model', 'Exllamav2HF'] or shared.args.loader in ['ExLlamav2', 'ExLlamav2_HF']:
|
16 |
+
add_lora_exllamav2(lora_names)
|
17 |
+
else:
|
18 |
+
add_lora_transformers(lora_names)
|
19 |
+
|
20 |
+
|
21 |
+
def get_lora_path(lora_name):
|
22 |
+
p = Path(lora_name)
|
23 |
+
if p.exists():
|
24 |
+
lora_name = p.parts[-1]
|
25 |
+
|
26 |
+
return Path(f"{shared.args.lora_dir}/{lora_name}")
|
27 |
+
|
28 |
+
|
29 |
+
def add_lora_exllamav2(lora_names):
|
30 |
+
|
31 |
+
from exllamav2 import ExLlamaV2Lora
|
32 |
+
|
33 |
+
if isinstance(shared.model.loras, list):
|
34 |
+
for lora in shared.model.loras:
|
35 |
+
lora.unload()
|
36 |
+
|
37 |
+
if len(lora_names) > 0:
|
38 |
+
logger.info("Applying the following LoRAs to {}: {}".format(shared.model_name, ', '.join(lora_names)))
|
39 |
+
shared.model.loras = []
|
40 |
+
for lora_name in lora_names:
|
41 |
+
lora_path = get_lora_path(lora_name)
|
42 |
+
if shared.model.__class__.__name__ == 'Exllamav2Model':
|
43 |
+
lora = ExLlamaV2Lora.from_directory(shared.model.model, str(lora_path))
|
44 |
+
else:
|
45 |
+
lora = ExLlamaV2Lora.from_directory(shared.model.ex_model, str(lora_path))
|
46 |
+
|
47 |
+
shared.model.loras.append(lora)
|
48 |
+
|
49 |
+
shared.lora_names = lora_names
|
50 |
+
else:
|
51 |
+
shared.lora_names = []
|
52 |
+
shared.model.loras = None
|
53 |
+
|
54 |
+
|
55 |
+
def add_lora_autogptq(lora_names):
|
56 |
+
'''
|
57 |
+
Adapted from https://github.com/Ph0rk0z/text-generation-webui-testing
|
58 |
+
'''
|
59 |
+
|
60 |
+
try:
|
61 |
+
from auto_gptq import get_gptq_peft_model
|
62 |
+
from auto_gptq.utils.peft_utils import GPTQLoraConfig
|
63 |
+
except:
|
64 |
+
logger.error("This version of AutoGPTQ does not support LoRA. You need to install from source or wait for a new release.")
|
65 |
+
return
|
66 |
+
|
67 |
+
if len(lora_names) == 0:
|
68 |
+
reload_model()
|
69 |
+
|
70 |
+
shared.lora_names = []
|
71 |
+
return
|
72 |
+
else:
|
73 |
+
if len(lora_names) > 1:
|
74 |
+
logger.warning('AutoGPTQ can only work with 1 LoRA at the moment. Only the first one in the list will be loaded.')
|
75 |
+
if not shared.args.no_inject_fused_attention:
|
76 |
+
logger.warning('Fused Atttention + AutoGPTQ may break Lora loading. Disable it.')
|
77 |
+
|
78 |
+
peft_config = GPTQLoraConfig(
|
79 |
+
inference_mode=True,
|
80 |
+
)
|
81 |
+
|
82 |
+
lora_path = get_lora_path(lora_names[0])
|
83 |
+
logger.info("Applying the following LoRAs to {}: {}".format(shared.model_name, ', '.join([lora_names[0]])))
|
84 |
+
shared.model = get_gptq_peft_model(shared.model, peft_config, lora_path)
|
85 |
+
shared.lora_names = [lora_names[0]]
|
86 |
+
return
|
87 |
+
|
88 |
+
|
89 |
+
def add_lora_transformers(lora_names):
|
90 |
+
prior_set = set(shared.lora_names)
|
91 |
+
added_set = set(lora_names) - prior_set
|
92 |
+
removed_set = prior_set - set(lora_names)
|
93 |
+
|
94 |
+
# If no LoRA needs to be added or removed, exit
|
95 |
+
if len(added_set) == 0 and len(removed_set) == 0:
|
96 |
+
return
|
97 |
+
|
98 |
+
# Add a LoRA when another LoRA is already present
|
99 |
+
if len(removed_set) == 0 and len(prior_set) > 0 and "__merged" not in shared.model.peft_config.keys():
|
100 |
+
logger.info(f"Adding the LoRA(s) named {added_set} to the model")
|
101 |
+
for lora in added_set:
|
102 |
+
shared.model.load_adapter(get_lora_path(lora), lora)
|
103 |
+
|
104 |
+
if len(lora_names) > 1:
|
105 |
+
merge_loras()
|
106 |
+
|
107 |
+
shared.lora_names = lora_names
|
108 |
+
return
|
109 |
+
|
110 |
+
# If any LoRA needs to be removed, start over
|
111 |
+
if len(removed_set) > 0:
|
112 |
+
shared.model = shared.model.unload()
|
113 |
+
|
114 |
+
if len(lora_names) > 0:
|
115 |
+
params = {}
|
116 |
+
if not shared.args.cpu:
|
117 |
+
if shared.args.load_in_4bit or shared.args.load_in_8bit:
|
118 |
+
params['peft_type'] = shared.model.dtype
|
119 |
+
else:
|
120 |
+
params['dtype'] = shared.model.dtype
|
121 |
+
if hasattr(shared.model, "hf_device_map"):
|
122 |
+
params['device_map'] = {"base_model.model." + k: v for k, v in shared.model.hf_device_map.items()}
|
123 |
+
|
124 |
+
logger.info("Applying the following LoRAs to {}: {}".format(shared.model_name, ', '.join(lora_names)))
|
125 |
+
shared.model = PeftModel.from_pretrained(shared.model, get_lora_path(lora_names[0]), adapter_name=lora_names[0], **params)
|
126 |
+
for lora in lora_names[1:]:
|
127 |
+
shared.model.load_adapter(get_lora_path(lora), lora)
|
128 |
+
|
129 |
+
if len(lora_names) > 1:
|
130 |
+
merge_loras()
|
131 |
+
|
132 |
+
if not shared.args.load_in_8bit and not shared.args.cpu:
|
133 |
+
shared.model.half()
|
134 |
+
if not hasattr(shared.model, "hf_device_map"):
|
135 |
+
if torch.backends.mps.is_available():
|
136 |
+
device = torch.device('mps')
|
137 |
+
shared.model = shared.model.to(device)
|
138 |
+
elif is_torch_xpu_available():
|
139 |
+
device = torch.device("xpu:0")
|
140 |
+
shared.model = shared.model.to(device)
|
141 |
+
else:
|
142 |
+
shared.model = shared.model.cuda()
|
143 |
+
|
144 |
+
shared.lora_names = lora_names
|
145 |
+
|
146 |
+
|
147 |
+
def merge_loras():
|
148 |
+
if len(list({shared.model.peft_config[adapter].r for adapter in shared.model.peft_config.keys()})) > 1:
|
149 |
+
logger.warning("The loaded LoRAs cannot be merged, as they have dissimilar ranks. Only the first one will be active.")
|
150 |
+
return
|
151 |
+
|
152 |
+
shared.model.add_weighted_adapter(shared.lora_names, [1] * len(shared.lora_names), "__merged")
|
153 |
+
shared.model.set_adapter("__merged")
|
modules/RoPE.py
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
def get_alpha_value(alpha, base):
|
2 |
+
'''
|
3 |
+
Gets alpha_value from alpha_value and rope_freq_base
|
4 |
+
'''
|
5 |
+
if base > 0:
|
6 |
+
return (base / 10000.) ** (63 / 64.)
|
7 |
+
else:
|
8 |
+
return alpha
|
9 |
+
|
10 |
+
|
11 |
+
def get_rope_freq_base(alpha, base):
|
12 |
+
'''
|
13 |
+
Gets rope_freq_base from alpha_value and rope_freq_base
|
14 |
+
'''
|
15 |
+
if base > 0:
|
16 |
+
return base
|
17 |
+
else:
|
18 |
+
return 10000 * alpha ** (64 / 63.)
|
modules/__pycache__/LoRA.cpython-311.pyc
ADDED
Binary file (9.93 kB). View file
|
|
modules/__pycache__/RoPE.cpython-311.pyc
ADDED
Binary file (733 Bytes). View file
|
|
modules/__pycache__/block_requests.cpython-311.pyc
ADDED
Binary file (3.11 kB). View file
|
|
modules/__pycache__/callbacks.cpython-311.pyc
ADDED
Binary file (5.88 kB). View file
|
|
modules/__pycache__/chat.cpython-311.pyc
ADDED
Binary file (46.9 kB). View file
|
|
modules/__pycache__/evaluate.cpython-311.pyc
ADDED
Binary file (8.69 kB). View file
|
|
modules/__pycache__/extensions.cpython-311.pyc
ADDED
Binary file (10.9 kB). View file
|
|
modules/__pycache__/github.cpython-311.pyc
ADDED
Binary file (2.25 kB). View file
|
|
modules/__pycache__/html_generator.cpython-311.pyc
ADDED
Binary file (15.8 kB). View file
|
|
modules/__pycache__/llamacpp_hf.cpython-311.pyc
ADDED
Binary file (12.7 kB). View file
|
|
modules/__pycache__/llamacpp_model.cpython-311.pyc
ADDED
Binary file (10.5 kB). View file
|
|
modules/__pycache__/loaders.cpython-311.pyc
ADDED
Binary file (7.04 kB). View file
|
|
modules/__pycache__/logging_colors.cpython-311.pyc
ADDED
Binary file (5.22 kB). View file
|
|
modules/__pycache__/logits.cpython-311.pyc
ADDED
Binary file (5.23 kB). View file
|
|
modules/__pycache__/metadata_gguf.cpython-311.pyc
ADDED
Binary file (4.74 kB). View file
|
|
modules/__pycache__/models.cpython-311.pyc
ADDED
Binary file (27 kB). View file
|
|
modules/__pycache__/models_settings.cpython-311.pyc
ADDED
Binary file (13.1 kB). View file
|
|
modules/__pycache__/one_click_installer_check.cpython-311.pyc
ADDED
Binary file (781 Bytes). View file
|
|
modules/__pycache__/presets.cpython-311.pyc
ADDED
Binary file (6.54 kB). View file
|
|
modules/__pycache__/prompts.cpython-311.pyc
ADDED
Binary file (1.33 kB). View file
|
|
modules/__pycache__/relative_imports.cpython-311.pyc
ADDED
Binary file (1.37 kB). View file
|
|
modules/__pycache__/sampler_hijack.cpython-311.pyc
ADDED
Binary file (24.7 kB). View file
|
|
modules/__pycache__/shared.cpython-311.pyc
ADDED
Binary file (26.6 kB). View file
|
|
modules/__pycache__/text_generation.cpython-311.pyc
ADDED
Binary file (25.2 kB). View file
|
|
modules/__pycache__/training.cpython-311.pyc
ADDED
Binary file (67.5 kB). View file
|
|
modules/__pycache__/ui.cpython-311.pyc
ADDED
Binary file (11.4 kB). View file
|
|
modules/__pycache__/ui_chat.cpython-311.pyc
ADDED
Binary file (59.1 kB). View file
|
|
modules/__pycache__/ui_default.cpython-311.pyc
ADDED
Binary file (16 kB). View file
|
|
modules/__pycache__/ui_file_saving.cpython-311.pyc
ADDED
Binary file (16.4 kB). View file
|
|
modules/__pycache__/ui_model_menu.cpython-311.pyc
ADDED
Binary file (38.4 kB). View file
|
|
modules/__pycache__/ui_notebook.cpython-311.pyc
ADDED
Binary file (16.7 kB). View file
|
|
modules/__pycache__/ui_parameters.cpython-311.pyc
ADDED
Binary file (21.5 kB). View file
|
|
modules/__pycache__/ui_session.cpython-311.pyc
ADDED
Binary file (9.11 kB). View file
|
|
modules/__pycache__/utils.cpython-311.pyc
ADDED
Binary file (13.5 kB). View file
|
|
modules/block_requests.py
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import builtins
|
2 |
+
import io
|
3 |
+
|
4 |
+
import requests
|
5 |
+
|
6 |
+
from modules.logging_colors import logger
|
7 |
+
|
8 |
+
original_open = open
|
9 |
+
original_get = requests.get
|
10 |
+
|
11 |
+
|
12 |
+
class RequestBlocker:
|
13 |
+
|
14 |
+
def __enter__(self):
|
15 |
+
requests.get = my_get
|
16 |
+
|
17 |
+
def __exit__(self, exc_type, exc_value, traceback):
|
18 |
+
requests.get = original_get
|
19 |
+
|
20 |
+
|
21 |
+
class OpenMonkeyPatch:
|
22 |
+
|
23 |
+
def __enter__(self):
|
24 |
+
builtins.open = my_open
|
25 |
+
|
26 |
+
def __exit__(self, exc_type, exc_value, traceback):
|
27 |
+
builtins.open = original_open
|
28 |
+
|
29 |
+
|
30 |
+
def my_get(url, **kwargs):
|
31 |
+
logger.info('Unwanted HTTP request redirected to localhost :)')
|
32 |
+
kwargs.setdefault('allow_redirects', True)
|
33 |
+
return requests.api.request('get', 'http://127.0.0.1/', **kwargs)
|
34 |
+
|
35 |
+
|
36 |
+
# Kindly provided by our friend WizardLM-30B
|
37 |
+
def my_open(*args, **kwargs):
|
38 |
+
filename = str(args[0])
|
39 |
+
if filename.endswith('index.html'):
|
40 |
+
with original_open(*args, **kwargs) as f:
|
41 |
+
file_contents = f.read()
|
42 |
+
|
43 |
+
file_contents = file_contents.replace(b'\t\t<script\n\t\t\tsrc="https://cdnjs.cloudflare.com/ajax/libs/iframe-resizer/4.3.7/iframeResizer.contentWindow.min.js"\n\t\t\tasync\n\t\t></script>', b'')
|
44 |
+
file_contents = file_contents.replace(b'cdnjs.cloudflare.com', b'127.0.0.1')
|
45 |
+
return io.BytesIO(file_contents)
|
46 |
+
else:
|
47 |
+
return original_open(*args, **kwargs)
|
modules/callbacks.py
ADDED
@@ -0,0 +1,103 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gc
|
2 |
+
import traceback
|
3 |
+
from queue import Queue
|
4 |
+
from threading import Thread
|
5 |
+
|
6 |
+
import torch
|
7 |
+
import transformers
|
8 |
+
from transformers import is_torch_xpu_available
|
9 |
+
|
10 |
+
import modules.shared as shared
|
11 |
+
|
12 |
+
|
13 |
+
class StopNowException(Exception):
|
14 |
+
pass
|
15 |
+
|
16 |
+
|
17 |
+
class _StopEverythingStoppingCriteria(transformers.StoppingCriteria):
|
18 |
+
def __init__(self):
|
19 |
+
transformers.StoppingCriteria.__init__(self)
|
20 |
+
|
21 |
+
def __call__(self, input_ids: torch.LongTensor, _scores: torch.FloatTensor) -> bool:
|
22 |
+
return shared.stop_everything
|
23 |
+
|
24 |
+
|
25 |
+
class Stream(transformers.StoppingCriteria):
|
26 |
+
def __init__(self, callback_func=None):
|
27 |
+
self.callback_func = callback_func
|
28 |
+
|
29 |
+
def __call__(self, input_ids, scores) -> bool:
|
30 |
+
if self.callback_func is not None:
|
31 |
+
self.callback_func(input_ids[0])
|
32 |
+
|
33 |
+
return False
|
34 |
+
|
35 |
+
|
36 |
+
class Iteratorize:
|
37 |
+
|
38 |
+
"""
|
39 |
+
Transforms a function that takes a callback
|
40 |
+
into a lazy iterator (generator).
|
41 |
+
|
42 |
+
Adapted from: https://stackoverflow.com/a/9969000
|
43 |
+
"""
|
44 |
+
|
45 |
+
def __init__(self, func, args=None, kwargs=None, callback=None):
|
46 |
+
self.mfunc = func
|
47 |
+
self.c_callback = callback
|
48 |
+
self.q = Queue()
|
49 |
+
self.sentinel = object()
|
50 |
+
self.args = args or []
|
51 |
+
self.kwargs = kwargs or {}
|
52 |
+
self.stop_now = False
|
53 |
+
|
54 |
+
def _callback(val):
|
55 |
+
if self.stop_now or shared.stop_everything:
|
56 |
+
raise StopNowException
|
57 |
+
self.q.put(val)
|
58 |
+
|
59 |
+
def gentask():
|
60 |
+
try:
|
61 |
+
ret = self.mfunc(callback=_callback, *args, **self.kwargs)
|
62 |
+
except StopNowException:
|
63 |
+
pass
|
64 |
+
except:
|
65 |
+
traceback.print_exc()
|
66 |
+
pass
|
67 |
+
|
68 |
+
clear_torch_cache()
|
69 |
+
self.q.put(self.sentinel)
|
70 |
+
if self.c_callback:
|
71 |
+
self.c_callback(ret)
|
72 |
+
|
73 |
+
self.thread = Thread(target=gentask)
|
74 |
+
self.thread.start()
|
75 |
+
|
76 |
+
def __iter__(self):
|
77 |
+
return self
|
78 |
+
|
79 |
+
def __next__(self):
|
80 |
+
obj = self.q.get(True, None)
|
81 |
+
if obj is self.sentinel:
|
82 |
+
raise StopIteration
|
83 |
+
else:
|
84 |
+
return obj
|
85 |
+
|
86 |
+
def __del__(self):
|
87 |
+
clear_torch_cache()
|
88 |
+
|
89 |
+
def __enter__(self):
|
90 |
+
return self
|
91 |
+
|
92 |
+
def __exit__(self, exc_type, exc_val, exc_tb):
|
93 |
+
self.stop_now = True
|
94 |
+
clear_torch_cache()
|
95 |
+
|
96 |
+
|
97 |
+
def clear_torch_cache():
|
98 |
+
gc.collect()
|
99 |
+
if not shared.args.cpu:
|
100 |
+
if is_torch_xpu_available():
|
101 |
+
torch.xpu.empty_cache()
|
102 |
+
else:
|
103 |
+
torch.cuda.empty_cache()
|
modules/chat.py
ADDED
@@ -0,0 +1,927 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import base64
|
2 |
+
import copy
|
3 |
+
import functools
|
4 |
+
import html
|
5 |
+
import json
|
6 |
+
import re
|
7 |
+
from datetime import datetime
|
8 |
+
from functools import partial
|
9 |
+
from pathlib import Path
|
10 |
+
|
11 |
+
import gradio as gr
|
12 |
+
import yaml
|
13 |
+
from jinja2.sandbox import ImmutableSandboxedEnvironment
|
14 |
+
from PIL import Image
|
15 |
+
|
16 |
+
import modules.shared as shared
|
17 |
+
from modules import utils
|
18 |
+
from modules.extensions import apply_extensions
|
19 |
+
from modules.html_generator import chat_html_wrapper, make_thumbnail
|
20 |
+
from modules.logging_colors import logger
|
21 |
+
from modules.text_generation import (
|
22 |
+
generate_reply,
|
23 |
+
get_encoded_length,
|
24 |
+
get_max_prompt_length
|
25 |
+
)
|
26 |
+
from modules.utils import delete_file, get_available_characters, save_file
|
27 |
+
|
28 |
+
# Copied from the Transformers library
|
29 |
+
jinja_env = ImmutableSandboxedEnvironment(trim_blocks=True, lstrip_blocks=True)
|
30 |
+
|
31 |
+
|
32 |
+
def str_presenter(dumper, data):
|
33 |
+
"""
|
34 |
+
Copied from https://github.com/yaml/pyyaml/issues/240
|
35 |
+
Makes pyyaml output prettier multiline strings.
|
36 |
+
"""
|
37 |
+
|
38 |
+
if data.count('\n') > 0:
|
39 |
+
return dumper.represent_scalar('tag:yaml.org,2002:str', data, style='|')
|
40 |
+
|
41 |
+
return dumper.represent_scalar('tag:yaml.org,2002:str', data)
|
42 |
+
|
43 |
+
|
44 |
+
yaml.add_representer(str, str_presenter)
|
45 |
+
yaml.representer.SafeRepresenter.add_representer(str, str_presenter)
|
46 |
+
|
47 |
+
|
48 |
+
def get_generation_prompt(renderer, impersonate=False, strip_trailing_spaces=True):
|
49 |
+
'''
|
50 |
+
Given a Jinja template, reverse-engineers the prefix and the suffix for
|
51 |
+
an assistant message (if impersonate=False) or an user message
|
52 |
+
(if impersonate=True)
|
53 |
+
'''
|
54 |
+
|
55 |
+
if impersonate:
|
56 |
+
messages = [
|
57 |
+
{"role": "user", "content": "<<|user-message-1|>>"},
|
58 |
+
{"role": "user", "content": "<<|user-message-2|>>"},
|
59 |
+
]
|
60 |
+
else:
|
61 |
+
messages = [
|
62 |
+
{"role": "assistant", "content": "<<|user-message-1|>>"},
|
63 |
+
{"role": "assistant", "content": "<<|user-message-2|>>"},
|
64 |
+
]
|
65 |
+
|
66 |
+
prompt = renderer(messages=messages)
|
67 |
+
|
68 |
+
suffix_plus_prefix = prompt.split("<<|user-message-1|>>")[1].split("<<|user-message-2|>>")[0]
|
69 |
+
suffix = prompt.split("<<|user-message-2|>>")[1]
|
70 |
+
prefix = suffix_plus_prefix[len(suffix):]
|
71 |
+
|
72 |
+
if strip_trailing_spaces:
|
73 |
+
prefix = prefix.rstrip(' ')
|
74 |
+
|
75 |
+
return prefix, suffix
|
76 |
+
|
77 |
+
|
78 |
+
def generate_chat_prompt(user_input, state, **kwargs):
|
79 |
+
impersonate = kwargs.get('impersonate', False)
|
80 |
+
_continue = kwargs.get('_continue', False)
|
81 |
+
also_return_rows = kwargs.get('also_return_rows', False)
|
82 |
+
history = kwargs.get('history', state['history'])['internal']
|
83 |
+
|
84 |
+
# Templates
|
85 |
+
chat_template = jinja_env.from_string(state['chat_template_str'])
|
86 |
+
instruction_template = jinja_env.from_string(state['instruction_template_str'])
|
87 |
+
chat_renderer = partial(chat_template.render, add_generation_prompt=False, name1=state['name1'], name2=state['name2'])
|
88 |
+
instruct_renderer = partial(instruction_template.render, add_generation_prompt=False)
|
89 |
+
|
90 |
+
messages = []
|
91 |
+
|
92 |
+
if state['mode'] == 'instruct':
|
93 |
+
renderer = instruct_renderer
|
94 |
+
if state['custom_system_message'].strip() != '':
|
95 |
+
messages.append({"role": "system", "content": state['custom_system_message']})
|
96 |
+
else:
|
97 |
+
renderer = chat_renderer
|
98 |
+
if state['context'].strip() != '':
|
99 |
+
context = replace_character_names(state['context'], state['name1'], state['name2'])
|
100 |
+
messages.append({"role": "system", "content": context})
|
101 |
+
|
102 |
+
insert_pos = len(messages)
|
103 |
+
for user_msg, assistant_msg in reversed(history):
|
104 |
+
user_msg = user_msg.strip()
|
105 |
+
assistant_msg = assistant_msg.strip()
|
106 |
+
|
107 |
+
if assistant_msg:
|
108 |
+
messages.insert(insert_pos, {"role": "assistant", "content": assistant_msg})
|
109 |
+
|
110 |
+
if user_msg not in ['', '<|BEGIN-VISIBLE-CHAT|>']:
|
111 |
+
messages.insert(insert_pos, {"role": "user", "content": user_msg})
|
112 |
+
|
113 |
+
user_input = user_input.strip()
|
114 |
+
if user_input and not impersonate and not _continue:
|
115 |
+
messages.append({"role": "user", "content": user_input})
|
116 |
+
|
117 |
+
def remove_extra_bos(prompt):
|
118 |
+
for bos_token in ['<s>', '<|startoftext|>']:
|
119 |
+
while prompt.startswith(bos_token):
|
120 |
+
prompt = prompt[len(bos_token):]
|
121 |
+
|
122 |
+
return prompt
|
123 |
+
|
124 |
+
def make_prompt(messages):
|
125 |
+
if state['mode'] == 'chat-instruct' and _continue:
|
126 |
+
prompt = renderer(messages=messages[:-1])
|
127 |
+
else:
|
128 |
+
prompt = renderer(messages=messages)
|
129 |
+
|
130 |
+
if state['mode'] == 'chat-instruct':
|
131 |
+
outer_messages = []
|
132 |
+
if state['custom_system_message'].strip() != '':
|
133 |
+
outer_messages.append({"role": "system", "content": state['custom_system_message']})
|
134 |
+
|
135 |
+
prompt = remove_extra_bos(prompt)
|
136 |
+
command = state['chat-instruct_command']
|
137 |
+
command = command.replace('<|character|>', state['name2'] if not impersonate else state['name1'])
|
138 |
+
command = command.replace('<|prompt|>', prompt)
|
139 |
+
|
140 |
+
if _continue:
|
141 |
+
prefix = get_generation_prompt(renderer, impersonate=impersonate, strip_trailing_spaces=False)[0]
|
142 |
+
prefix += messages[-1]["content"]
|
143 |
+
else:
|
144 |
+
prefix = get_generation_prompt(renderer, impersonate=impersonate)[0]
|
145 |
+
if not impersonate:
|
146 |
+
prefix = apply_extensions('bot_prefix', prefix, state)
|
147 |
+
|
148 |
+
outer_messages.append({"role": "user", "content": command})
|
149 |
+
outer_messages.append({"role": "assistant", "content": prefix})
|
150 |
+
|
151 |
+
prompt = instruction_template.render(messages=outer_messages)
|
152 |
+
suffix = get_generation_prompt(instruct_renderer, impersonate=False)[1]
|
153 |
+
prompt = prompt[:-len(suffix)]
|
154 |
+
|
155 |
+
else:
|
156 |
+
if _continue:
|
157 |
+
suffix = get_generation_prompt(renderer, impersonate=impersonate)[1]
|
158 |
+
prompt = prompt[:-len(suffix)]
|
159 |
+
else:
|
160 |
+
prefix = get_generation_prompt(renderer, impersonate=impersonate)[0]
|
161 |
+
if state['mode'] == 'chat' and not impersonate:
|
162 |
+
prefix = apply_extensions('bot_prefix', prefix, state)
|
163 |
+
|
164 |
+
prompt += prefix
|
165 |
+
|
166 |
+
prompt = remove_extra_bos(prompt)
|
167 |
+
return prompt
|
168 |
+
|
169 |
+
# Handle truncation
|
170 |
+
max_length = get_max_prompt_length(state)
|
171 |
+
prompt = make_prompt(messages)
|
172 |
+
encoded_length = get_encoded_length(prompt)
|
173 |
+
|
174 |
+
while len(messages) > 0 and encoded_length > max_length:
|
175 |
+
|
176 |
+
# Remove old message, save system message
|
177 |
+
if len(messages) > 2 and messages[0]['role'] == 'system':
|
178 |
+
messages.pop(1)
|
179 |
+
|
180 |
+
# Remove old message when no system message is present
|
181 |
+
elif len(messages) > 1 and messages[0]['role'] != 'system':
|
182 |
+
messages.pop(0)
|
183 |
+
|
184 |
+
# Resort to truncating the user input
|
185 |
+
else:
|
186 |
+
|
187 |
+
user_message = messages[-1]['content']
|
188 |
+
|
189 |
+
# Bisect the truncation point
|
190 |
+
left, right = 0, len(user_message) - 1
|
191 |
+
|
192 |
+
while right - left > 1:
|
193 |
+
mid = (left + right) // 2
|
194 |
+
|
195 |
+
messages[-1]['content'] = user_message[mid:]
|
196 |
+
prompt = make_prompt(messages)
|
197 |
+
encoded_length = get_encoded_length(prompt)
|
198 |
+
|
199 |
+
if encoded_length <= max_length:
|
200 |
+
right = mid
|
201 |
+
else:
|
202 |
+
left = mid
|
203 |
+
|
204 |
+
messages[-1]['content'] = user_message[right:]
|
205 |
+
prompt = make_prompt(messages)
|
206 |
+
encoded_length = get_encoded_length(prompt)
|
207 |
+
if encoded_length > max_length:
|
208 |
+
logger.error(f"Failed to build the chat prompt. The input is too long for the available context length.\n\nTruncation length: {state['truncation_length']}\nmax_new_tokens: {state['max_new_tokens']} (is it too high?)\nAvailable context length: {max_length}\n")
|
209 |
+
raise ValueError
|
210 |
+
else:
|
211 |
+
logger.warning(f"The input has been truncated. Context length: {state['truncation_length']}, max_new_tokens: {state['max_new_tokens']}, available context length: {max_length}.")
|
212 |
+
break
|
213 |
+
|
214 |
+
prompt = make_prompt(messages)
|
215 |
+
encoded_length = get_encoded_length(prompt)
|
216 |
+
|
217 |
+
if also_return_rows:
|
218 |
+
return prompt, [message['content'] for message in messages]
|
219 |
+
else:
|
220 |
+
return prompt
|
221 |
+
|
222 |
+
|
223 |
+
def get_stopping_strings(state):
|
224 |
+
stopping_strings = []
|
225 |
+
renderers = []
|
226 |
+
|
227 |
+
if state['mode'] in ['instruct', 'chat-instruct']:
|
228 |
+
template = jinja_env.from_string(state['instruction_template_str'])
|
229 |
+
renderer = partial(template.render, add_generation_prompt=False)
|
230 |
+
renderers.append(renderer)
|
231 |
+
|
232 |
+
if state['mode'] in ['chat', 'chat-instruct']:
|
233 |
+
template = jinja_env.from_string(state['chat_template_str'])
|
234 |
+
renderer = partial(template.render, add_generation_prompt=False, name1=state['name1'], name2=state['name2'])
|
235 |
+
renderers.append(renderer)
|
236 |
+
|
237 |
+
for renderer in renderers:
|
238 |
+
prefix_bot, suffix_bot = get_generation_prompt(renderer, impersonate=False)
|
239 |
+
prefix_user, suffix_user = get_generation_prompt(renderer, impersonate=True)
|
240 |
+
|
241 |
+
stopping_strings += [
|
242 |
+
suffix_user + prefix_bot,
|
243 |
+
suffix_user + prefix_user,
|
244 |
+
suffix_bot + prefix_bot,
|
245 |
+
suffix_bot + prefix_user,
|
246 |
+
]
|
247 |
+
|
248 |
+
if 'stopping_strings' in state and isinstance(state['stopping_strings'], list):
|
249 |
+
stopping_strings += state.pop('stopping_strings')
|
250 |
+
|
251 |
+
return list(set(stopping_strings))
|
252 |
+
|
253 |
+
|
254 |
+
def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_message=True, for_ui=False):
|
255 |
+
history = state['history']
|
256 |
+
output = copy.deepcopy(history)
|
257 |
+
output = apply_extensions('history', output)
|
258 |
+
state = apply_extensions('state', state)
|
259 |
+
|
260 |
+
visible_text = None
|
261 |
+
stopping_strings = get_stopping_strings(state)
|
262 |
+
is_stream = state['stream']
|
263 |
+
|
264 |
+
# Prepare the input
|
265 |
+
if not (regenerate or _continue):
|
266 |
+
visible_text = html.escape(text)
|
267 |
+
|
268 |
+
# Apply extensions
|
269 |
+
text, visible_text = apply_extensions('chat_input', text, visible_text, state)
|
270 |
+
text = apply_extensions('input', text, state, is_chat=True)
|
271 |
+
|
272 |
+
output['internal'].append([text, ''])
|
273 |
+
output['visible'].append([visible_text, ''])
|
274 |
+
|
275 |
+
# *Is typing...*
|
276 |
+
if loading_message:
|
277 |
+
yield {
|
278 |
+
'visible': output['visible'][:-1] + [[output['visible'][-1][0], shared.processing_message]],
|
279 |
+
'internal': output['internal']
|
280 |
+
}
|
281 |
+
else:
|
282 |
+
text, visible_text = output['internal'][-1][0], output['visible'][-1][0]
|
283 |
+
if regenerate:
|
284 |
+
if loading_message:
|
285 |
+
yield {
|
286 |
+
'visible': output['visible'][:-1] + [[visible_text, shared.processing_message]],
|
287 |
+
'internal': output['internal'][:-1] + [[text, '']]
|
288 |
+
}
|
289 |
+
elif _continue:
|
290 |
+
last_reply = [output['internal'][-1][1], output['visible'][-1][1]]
|
291 |
+
if loading_message:
|
292 |
+
yield {
|
293 |
+
'visible': output['visible'][:-1] + [[visible_text, last_reply[1] + '...']],
|
294 |
+
'internal': output['internal']
|
295 |
+
}
|
296 |
+
|
297 |
+
if shared.model_name == 'None' or shared.model is None:
|
298 |
+
raise ValueError("No model is loaded! Select one in the Model tab.")
|
299 |
+
|
300 |
+
# Generate the prompt
|
301 |
+
kwargs = {
|
302 |
+
'_continue': _continue,
|
303 |
+
'history': output if _continue else {k: v[:-1] for k, v in output.items()}
|
304 |
+
}
|
305 |
+
prompt = apply_extensions('custom_generate_chat_prompt', text, state, **kwargs)
|
306 |
+
if prompt is None:
|
307 |
+
prompt = generate_chat_prompt(text, state, **kwargs)
|
308 |
+
|
309 |
+
# Generate
|
310 |
+
reply = None
|
311 |
+
for j, reply in enumerate(generate_reply(prompt, state, stopping_strings=stopping_strings, is_chat=True, for_ui=for_ui)):
|
312 |
+
|
313 |
+
# Extract the reply
|
314 |
+
visible_reply = reply
|
315 |
+
if state['mode'] in ['chat', 'chat-instruct']:
|
316 |
+
visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
|
317 |
+
|
318 |
+
visible_reply = html.escape(visible_reply)
|
319 |
+
|
320 |
+
if shared.stop_everything:
|
321 |
+
output['visible'][-1][1] = apply_extensions('output', output['visible'][-1][1], state, is_chat=True)
|
322 |
+
yield output
|
323 |
+
return
|
324 |
+
|
325 |
+
if _continue:
|
326 |
+
output['internal'][-1] = [text, last_reply[0] + reply]
|
327 |
+
output['visible'][-1] = [visible_text, last_reply[1] + visible_reply]
|
328 |
+
if is_stream:
|
329 |
+
yield output
|
330 |
+
elif not (j == 0 and visible_reply.strip() == ''):
|
331 |
+
output['internal'][-1] = [text, reply.lstrip(' ')]
|
332 |
+
output['visible'][-1] = [visible_text, visible_reply.lstrip(' ')]
|
333 |
+
if is_stream:
|
334 |
+
yield output
|
335 |
+
|
336 |
+
output['visible'][-1][1] = apply_extensions('output', output['visible'][-1][1], state, is_chat=True)
|
337 |
+
yield output
|
338 |
+
|
339 |
+
|
340 |
+
def impersonate_wrapper(text, state):
|
341 |
+
|
342 |
+
static_output = chat_html_wrapper(state['history'], state['name1'], state['name2'], state['mode'], state['chat_style'], state['character_menu'])
|
343 |
+
|
344 |
+
if shared.model_name == 'None' or shared.model is None:
|
345 |
+
logger.error("No model is loaded! Select one in the Model tab.")
|
346 |
+
yield '', static_output
|
347 |
+
return
|
348 |
+
|
349 |
+
prompt = generate_chat_prompt('', state, impersonate=True)
|
350 |
+
stopping_strings = get_stopping_strings(state)
|
351 |
+
|
352 |
+
yield text + '...', static_output
|
353 |
+
reply = None
|
354 |
+
for reply in generate_reply(prompt + text, state, stopping_strings=stopping_strings, is_chat=True):
|
355 |
+
yield (text + reply).lstrip(' '), static_output
|
356 |
+
if shared.stop_everything:
|
357 |
+
return
|
358 |
+
|
359 |
+
|
360 |
+
def generate_chat_reply(text, state, regenerate=False, _continue=False, loading_message=True, for_ui=False):
|
361 |
+
history = state['history']
|
362 |
+
if regenerate or _continue:
|
363 |
+
text = ''
|
364 |
+
if (len(history['visible']) == 1 and not history['visible'][0][0]) or len(history['internal']) == 0:
|
365 |
+
yield history
|
366 |
+
return
|
367 |
+
|
368 |
+
for history in chatbot_wrapper(text, state, regenerate=regenerate, _continue=_continue, loading_message=loading_message, for_ui=for_ui):
|
369 |
+
yield history
|
370 |
+
|
371 |
+
|
372 |
+
def character_is_loaded(state, raise_exception=False):
|
373 |
+
if state['mode'] in ['chat', 'chat-instruct'] and state['name2'] == '':
|
374 |
+
logger.error('It looks like no character is loaded. Please load one under Parameters > Character.')
|
375 |
+
if raise_exception:
|
376 |
+
raise ValueError
|
377 |
+
|
378 |
+
return False
|
379 |
+
else:
|
380 |
+
return True
|
381 |
+
|
382 |
+
|
383 |
+
def generate_chat_reply_wrapper(text, state, regenerate=False, _continue=False):
|
384 |
+
'''
|
385 |
+
Same as above but returns HTML for the UI
|
386 |
+
'''
|
387 |
+
|
388 |
+
if not character_is_loaded(state):
|
389 |
+
return
|
390 |
+
|
391 |
+
if state['start_with'] != '' and not _continue:
|
392 |
+
if regenerate:
|
393 |
+
text, state['history'] = remove_last_message(state['history'])
|
394 |
+
regenerate = False
|
395 |
+
|
396 |
+
_continue = True
|
397 |
+
send_dummy_message(text, state)
|
398 |
+
send_dummy_reply(state['start_with'], state)
|
399 |
+
|
400 |
+
for i, history in enumerate(generate_chat_reply(text, state, regenerate, _continue, loading_message=True, for_ui=True)):
|
401 |
+
yield chat_html_wrapper(history, state['name1'], state['name2'], state['mode'], state['chat_style'], state['character_menu']), history
|
402 |
+
|
403 |
+
|
404 |
+
def remove_last_message(history):
|
405 |
+
if len(history['visible']) > 0 and history['internal'][-1][0] != '<|BEGIN-VISIBLE-CHAT|>':
|
406 |
+
last = history['visible'].pop()
|
407 |
+
history['internal'].pop()
|
408 |
+
else:
|
409 |
+
last = ['', '']
|
410 |
+
|
411 |
+
return html.unescape(last[0]), history
|
412 |
+
|
413 |
+
|
414 |
+
def send_last_reply_to_input(history):
|
415 |
+
if len(history['visible']) > 0:
|
416 |
+
return html.unescape(history['visible'][-1][1])
|
417 |
+
else:
|
418 |
+
return ''
|
419 |
+
|
420 |
+
|
421 |
+
def replace_last_reply(text, state):
|
422 |
+
history = state['history']
|
423 |
+
|
424 |
+
if len(text.strip()) == 0:
|
425 |
+
return history
|
426 |
+
elif len(history['visible']) > 0:
|
427 |
+
history['visible'][-1][1] = html.escape(text)
|
428 |
+
history['internal'][-1][1] = apply_extensions('input', text, state, is_chat=True)
|
429 |
+
|
430 |
+
return history
|
431 |
+
|
432 |
+
|
433 |
+
def send_dummy_message(text, state):
|
434 |
+
history = state['history']
|
435 |
+
history['visible'].append([html.escape(text), ''])
|
436 |
+
history['internal'].append([apply_extensions('input', text, state, is_chat=True), ''])
|
437 |
+
return history
|
438 |
+
|
439 |
+
|
440 |
+
def send_dummy_reply(text, state):
|
441 |
+
history = state['history']
|
442 |
+
if len(history['visible']) > 0 and not history['visible'][-1][1] == '':
|
443 |
+
history['visible'].append(['', ''])
|
444 |
+
history['internal'].append(['', ''])
|
445 |
+
|
446 |
+
history['visible'][-1][1] = html.escape(text)
|
447 |
+
history['internal'][-1][1] = apply_extensions('input', text, state, is_chat=True)
|
448 |
+
return history
|
449 |
+
|
450 |
+
|
451 |
+
def redraw_html(history, name1, name2, mode, style, character, reset_cache=False):
|
452 |
+
return chat_html_wrapper(history, name1, name2, mode, style, character, reset_cache=reset_cache)
|
453 |
+
|
454 |
+
|
455 |
+
def start_new_chat(state):
|
456 |
+
mode = state['mode']
|
457 |
+
history = {'internal': [], 'visible': []}
|
458 |
+
|
459 |
+
if mode != 'instruct':
|
460 |
+
greeting = replace_character_names(state['greeting'], state['name1'], state['name2'])
|
461 |
+
if greeting != '':
|
462 |
+
history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
|
463 |
+
history['visible'] += [['', apply_extensions('output', greeting, state, is_chat=True)]]
|
464 |
+
|
465 |
+
unique_id = datetime.now().strftime('%Y%m%d-%H-%M-%S')
|
466 |
+
save_history(history, unique_id, state['character_menu'], state['mode'])
|
467 |
+
|
468 |
+
return history
|
469 |
+
|
470 |
+
|
471 |
+
def get_history_file_path(unique_id, character, mode):
|
472 |
+
if mode == 'instruct':
|
473 |
+
p = Path(f'logs/instruct/{unique_id}.json')
|
474 |
+
else:
|
475 |
+
p = Path(f'logs/chat/{character}/{unique_id}.json')
|
476 |
+
|
477 |
+
return p
|
478 |
+
|
479 |
+
|
480 |
+
def save_history(history, unique_id, character, mode):
|
481 |
+
if shared.args.multi_user:
|
482 |
+
return
|
483 |
+
|
484 |
+
p = get_history_file_path(unique_id, character, mode)
|
485 |
+
if not p.parent.is_dir():
|
486 |
+
p.parent.mkdir(parents=True)
|
487 |
+
|
488 |
+
with open(p, 'w', encoding='utf-8') as f:
|
489 |
+
f.write(json.dumps(history, indent=4))
|
490 |
+
|
491 |
+
|
492 |
+
def rename_history(old_id, new_id, character, mode):
|
493 |
+
if shared.args.multi_user:
|
494 |
+
return
|
495 |
+
|
496 |
+
old_p = get_history_file_path(old_id, character, mode)
|
497 |
+
new_p = get_history_file_path(new_id, character, mode)
|
498 |
+
if new_p.parent != old_p.parent:
|
499 |
+
logger.error(f"The following path is not allowed: {new_p}.")
|
500 |
+
elif new_p == old_p:
|
501 |
+
logger.info("The provided path is identical to the old one.")
|
502 |
+
else:
|
503 |
+
logger.info(f"Renaming {old_p} to {new_p}")
|
504 |
+
old_p.rename(new_p)
|
505 |
+
|
506 |
+
|
507 |
+
def find_all_histories(state):
|
508 |
+
if shared.args.multi_user:
|
509 |
+
return ['']
|
510 |
+
|
511 |
+
if state['mode'] == 'instruct':
|
512 |
+
paths = Path('logs/instruct').glob('*.json')
|
513 |
+
else:
|
514 |
+
character = state['character_menu']
|
515 |
+
|
516 |
+
# Handle obsolete filenames and paths
|
517 |
+
old_p = Path(f'logs/{character}_persistent.json')
|
518 |
+
new_p = Path(f'logs/persistent_{character}.json')
|
519 |
+
if old_p.exists():
|
520 |
+
logger.warning(f"Renaming {old_p} to {new_p}")
|
521 |
+
old_p.rename(new_p)
|
522 |
+
if new_p.exists():
|
523 |
+
unique_id = datetime.now().strftime('%Y%m%d-%H-%M-%S')
|
524 |
+
p = get_history_file_path(unique_id, character, state['mode'])
|
525 |
+
logger.warning(f"Moving {new_p} to {p}")
|
526 |
+
p.parent.mkdir(exist_ok=True)
|
527 |
+
new_p.rename(p)
|
528 |
+
|
529 |
+
paths = Path(f'logs/chat/{character}').glob('*.json')
|
530 |
+
|
531 |
+
histories = sorted(paths, key=lambda x: x.stat().st_mtime, reverse=True)
|
532 |
+
histories = [path.stem for path in histories]
|
533 |
+
|
534 |
+
return histories
|
535 |
+
|
536 |
+
|
537 |
+
def load_latest_history(state):
|
538 |
+
'''
|
539 |
+
Loads the latest history for the given character in chat or chat-instruct
|
540 |
+
mode, or the latest instruct history for instruct mode.
|
541 |
+
'''
|
542 |
+
|
543 |
+
if shared.args.multi_user:
|
544 |
+
return start_new_chat(state)
|
545 |
+
|
546 |
+
histories = find_all_histories(state)
|
547 |
+
|
548 |
+
if len(histories) > 0:
|
549 |
+
history = load_history(histories[0], state['character_menu'], state['mode'])
|
550 |
+
else:
|
551 |
+
history = start_new_chat(state)
|
552 |
+
|
553 |
+
return history
|
554 |
+
|
555 |
+
|
556 |
+
def load_history_after_deletion(state, idx):
|
557 |
+
'''
|
558 |
+
Loads the latest history for the given character in chat or chat-instruct
|
559 |
+
mode, or the latest instruct history for instruct mode.
|
560 |
+
'''
|
561 |
+
|
562 |
+
if shared.args.multi_user:
|
563 |
+
return start_new_chat(state)
|
564 |
+
|
565 |
+
histories = find_all_histories(state)
|
566 |
+
idx = min(int(idx), len(histories) - 1)
|
567 |
+
idx = max(0, idx)
|
568 |
+
|
569 |
+
if len(histories) > 0:
|
570 |
+
history = load_history(histories[idx], state['character_menu'], state['mode'])
|
571 |
+
else:
|
572 |
+
history = start_new_chat(state)
|
573 |
+
histories = find_all_histories(state)
|
574 |
+
|
575 |
+
return history, gr.update(choices=histories, value=histories[idx])
|
576 |
+
|
577 |
+
|
578 |
+
def update_character_menu_after_deletion(idx):
|
579 |
+
characters = utils.get_available_characters()
|
580 |
+
idx = min(int(idx), len(characters) - 1)
|
581 |
+
idx = max(0, idx)
|
582 |
+
return gr.update(choices=characters, value=characters[idx])
|
583 |
+
|
584 |
+
|
585 |
+
def load_history(unique_id, character, mode):
|
586 |
+
p = get_history_file_path(unique_id, character, mode)
|
587 |
+
|
588 |
+
f = json.loads(open(p, 'rb').read())
|
589 |
+
if 'internal' in f and 'visible' in f:
|
590 |
+
history = f
|
591 |
+
else:
|
592 |
+
history = {
|
593 |
+
'internal': f['data'],
|
594 |
+
'visible': f['data_visible']
|
595 |
+
}
|
596 |
+
|
597 |
+
return history
|
598 |
+
|
599 |
+
|
600 |
+
def load_history_json(file, history):
|
601 |
+
try:
|
602 |
+
file = file.decode('utf-8')
|
603 |
+
f = json.loads(file)
|
604 |
+
if 'internal' in f and 'visible' in f:
|
605 |
+
history = f
|
606 |
+
else:
|
607 |
+
history = {
|
608 |
+
'internal': f['data'],
|
609 |
+
'visible': f['data_visible']
|
610 |
+
}
|
611 |
+
|
612 |
+
return history
|
613 |
+
except:
|
614 |
+
return history
|
615 |
+
|
616 |
+
|
617 |
+
def delete_history(unique_id, character, mode):
|
618 |
+
p = get_history_file_path(unique_id, character, mode)
|
619 |
+
delete_file(p)
|
620 |
+
|
621 |
+
|
622 |
+
def replace_character_names(text, name1, name2):
|
623 |
+
text = text.replace('{{user}}', name1).replace('{{char}}', name2)
|
624 |
+
return text.replace('<USER>', name1).replace('<BOT>', name2)
|
625 |
+
|
626 |
+
|
627 |
+
def generate_pfp_cache(character):
|
628 |
+
cache_folder = Path(shared.args.disk_cache_dir)
|
629 |
+
if not cache_folder.exists():
|
630 |
+
cache_folder.mkdir()
|
631 |
+
|
632 |
+
for path in [Path(f"characters/{character}.{extension}") for extension in ['png', 'jpg', 'jpeg']]:
|
633 |
+
if path.exists():
|
634 |
+
original_img = Image.open(path)
|
635 |
+
original_img.save(Path(f'{cache_folder}/pfp_character.png'), format='PNG')
|
636 |
+
|
637 |
+
thumb = make_thumbnail(original_img)
|
638 |
+
thumb.save(Path(f'{cache_folder}/pfp_character_thumb.png'), format='PNG')
|
639 |
+
|
640 |
+
return thumb
|
641 |
+
|
642 |
+
return None
|
643 |
+
|
644 |
+
|
645 |
+
def load_character(character, name1, name2):
|
646 |
+
context = greeting = ""
|
647 |
+
greeting_field = 'greeting'
|
648 |
+
picture = None
|
649 |
+
|
650 |
+
filepath = None
|
651 |
+
for extension in ["yml", "yaml", "json"]:
|
652 |
+
filepath = Path(f'characters/{character}.{extension}')
|
653 |
+
if filepath.exists():
|
654 |
+
break
|
655 |
+
|
656 |
+
if filepath is None or not filepath.exists():
|
657 |
+
logger.error(f"Could not find the character \"{character}\" inside characters/. No character has been loaded.")
|
658 |
+
raise ValueError
|
659 |
+
|
660 |
+
file_contents = open(filepath, 'r', encoding='utf-8').read()
|
661 |
+
data = json.loads(file_contents) if extension == "json" else yaml.safe_load(file_contents)
|
662 |
+
cache_folder = Path(shared.args.disk_cache_dir)
|
663 |
+
|
664 |
+
for path in [Path(f"{cache_folder}/pfp_character.png"), Path(f"{cache_folder}/pfp_character_thumb.png")]:
|
665 |
+
if path.exists():
|
666 |
+
path.unlink()
|
667 |
+
|
668 |
+
picture = generate_pfp_cache(character)
|
669 |
+
|
670 |
+
# Finding the bot's name
|
671 |
+
for k in ['name', 'bot', '<|bot|>', 'char_name']:
|
672 |
+
if k in data and data[k] != '':
|
673 |
+
name2 = data[k]
|
674 |
+
break
|
675 |
+
|
676 |
+
# Find the user name (if any)
|
677 |
+
for k in ['your_name', 'user', '<|user|>']:
|
678 |
+
if k in data and data[k] != '':
|
679 |
+
name1 = data[k]
|
680 |
+
break
|
681 |
+
|
682 |
+
if 'context' in data:
|
683 |
+
context = data['context'].strip()
|
684 |
+
elif "char_persona" in data:
|
685 |
+
context = build_pygmalion_style_context(data)
|
686 |
+
greeting_field = 'char_greeting'
|
687 |
+
|
688 |
+
greeting = data.get(greeting_field, greeting)
|
689 |
+
return name1, name2, picture, greeting, context
|
690 |
+
|
691 |
+
|
692 |
+
def load_instruction_template(template):
|
693 |
+
for filepath in [Path(f'instruction-templates/{template}.yaml'), Path('instruction-templates/Alpaca.yaml')]:
|
694 |
+
if filepath.exists():
|
695 |
+
break
|
696 |
+
else:
|
697 |
+
return ''
|
698 |
+
|
699 |
+
file_contents = open(filepath, 'r', encoding='utf-8').read()
|
700 |
+
data = yaml.safe_load(file_contents)
|
701 |
+
if 'instruction_template' in data:
|
702 |
+
return data['instruction_template']
|
703 |
+
else:
|
704 |
+
return jinja_template_from_old_format(data)
|
705 |
+
|
706 |
+
|
707 |
+
@functools.cache
|
708 |
+
def load_character_memoized(character, name1, name2):
|
709 |
+
return load_character(character, name1, name2)
|
710 |
+
|
711 |
+
|
712 |
+
@functools.cache
|
713 |
+
def load_instruction_template_memoized(template):
|
714 |
+
return load_instruction_template(template)
|
715 |
+
|
716 |
+
|
717 |
+
def upload_character(file, img, tavern=False):
|
718 |
+
decoded_file = file if isinstance(file, str) else file.decode('utf-8')
|
719 |
+
try:
|
720 |
+
data = json.loads(decoded_file)
|
721 |
+
except:
|
722 |
+
data = yaml.safe_load(decoded_file)
|
723 |
+
|
724 |
+
if 'char_name' in data:
|
725 |
+
name = data['char_name']
|
726 |
+
greeting = data['char_greeting']
|
727 |
+
context = build_pygmalion_style_context(data)
|
728 |
+
yaml_data = generate_character_yaml(name, greeting, context)
|
729 |
+
else:
|
730 |
+
name = data['name']
|
731 |
+
yaml_data = generate_character_yaml(data['name'], data['greeting'], data['context'])
|
732 |
+
|
733 |
+
outfile_name = name
|
734 |
+
i = 1
|
735 |
+
while Path(f'characters/{outfile_name}.yaml').exists():
|
736 |
+
outfile_name = f'{name}_{i:03d}'
|
737 |
+
i += 1
|
738 |
+
|
739 |
+
with open(Path(f'characters/{outfile_name}.yaml'), 'w', encoding='utf-8') as f:
|
740 |
+
f.write(yaml_data)
|
741 |
+
|
742 |
+
if img is not None:
|
743 |
+
img.save(Path(f'characters/{outfile_name}.png'))
|
744 |
+
|
745 |
+
logger.info(f'New character saved to "characters/{outfile_name}.yaml".')
|
746 |
+
return gr.update(value=outfile_name, choices=get_available_characters())
|
747 |
+
|
748 |
+
|
749 |
+
def build_pygmalion_style_context(data):
|
750 |
+
context = ""
|
751 |
+
if 'char_persona' in data and data['char_persona'] != '':
|
752 |
+
context += f"{data['char_name']}'s Persona: {data['char_persona']}\n"
|
753 |
+
|
754 |
+
if 'world_scenario' in data and data['world_scenario'] != '':
|
755 |
+
context += f"Scenario: {data['world_scenario']}\n"
|
756 |
+
|
757 |
+
if 'example_dialogue' in data and data['example_dialogue'] != '':
|
758 |
+
context += f"{data['example_dialogue'].strip()}\n"
|
759 |
+
|
760 |
+
context = f"{context.strip()}\n"
|
761 |
+
return context
|
762 |
+
|
763 |
+
|
764 |
+
def upload_tavern_character(img, _json):
|
765 |
+
_json = {'char_name': _json['name'], 'char_persona': _json['description'], 'char_greeting': _json['first_mes'], 'example_dialogue': _json['mes_example'], 'world_scenario': _json['scenario']}
|
766 |
+
return upload_character(json.dumps(_json), img, tavern=True)
|
767 |
+
|
768 |
+
|
769 |
+
def check_tavern_character(img):
|
770 |
+
if "chara" not in img.info:
|
771 |
+
return "Not a TavernAI card", None, None, gr.update(interactive=False)
|
772 |
+
|
773 |
+
decoded_string = base64.b64decode(img.info['chara']).replace(b'\\r\\n', b'\\n')
|
774 |
+
_json = json.loads(decoded_string)
|
775 |
+
if "data" in _json:
|
776 |
+
_json = _json["data"]
|
777 |
+
|
778 |
+
return _json['name'], _json['description'], _json, gr.update(interactive=True)
|
779 |
+
|
780 |
+
|
781 |
+
def upload_your_profile_picture(img):
|
782 |
+
cache_folder = Path(shared.args.disk_cache_dir)
|
783 |
+
if not cache_folder.exists():
|
784 |
+
cache_folder.mkdir()
|
785 |
+
|
786 |
+
if img is None:
|
787 |
+
if Path(f"{cache_folder}/pfp_me.png").exists():
|
788 |
+
Path(f"{cache_folder}/pfp_me.png").unlink()
|
789 |
+
else:
|
790 |
+
img = make_thumbnail(img)
|
791 |
+
img.save(Path(f'{cache_folder}/pfp_me.png'))
|
792 |
+
logger.info(f'Profile picture saved to "{cache_folder}/pfp_me.png"')
|
793 |
+
|
794 |
+
|
795 |
+
def generate_character_yaml(name, greeting, context):
|
796 |
+
data = {
|
797 |
+
'name': name,
|
798 |
+
'greeting': greeting,
|
799 |
+
'context': context,
|
800 |
+
}
|
801 |
+
|
802 |
+
data = {k: v for k, v in data.items() if v} # Strip falsy
|
803 |
+
return yaml.dump(data, sort_keys=False, width=float("inf"))
|
804 |
+
|
805 |
+
|
806 |
+
def generate_instruction_template_yaml(instruction_template):
|
807 |
+
data = {
|
808 |
+
'instruction_template': instruction_template
|
809 |
+
}
|
810 |
+
|
811 |
+
return my_yaml_output(data)
|
812 |
+
|
813 |
+
|
814 |
+
def save_character(name, greeting, context, picture, filename):
|
815 |
+
if filename == "":
|
816 |
+
logger.error("The filename is empty, so the character will not be saved.")
|
817 |
+
return
|
818 |
+
|
819 |
+
data = generate_character_yaml(name, greeting, context)
|
820 |
+
filepath = Path(f'characters/{filename}.yaml')
|
821 |
+
save_file(filepath, data)
|
822 |
+
path_to_img = Path(f'characters/{filename}.png')
|
823 |
+
if picture is not None:
|
824 |
+
picture.save(path_to_img)
|
825 |
+
logger.info(f'Saved {path_to_img}.')
|
826 |
+
|
827 |
+
|
828 |
+
def delete_character(name, instruct=False):
|
829 |
+
for extension in ["yml", "yaml", "json"]:
|
830 |
+
delete_file(Path(f'characters/{name}.{extension}'))
|
831 |
+
|
832 |
+
delete_file(Path(f'characters/{name}.png'))
|
833 |
+
|
834 |
+
|
835 |
+
def jinja_template_from_old_format(params, verbose=False):
|
836 |
+
MASTER_TEMPLATE = """
|
837 |
+
{%- set ns = namespace(found=false) -%}
|
838 |
+
{%- for message in messages -%}
|
839 |
+
{%- if message['role'] == 'system' -%}
|
840 |
+
{%- set ns.found = true -%}
|
841 |
+
{%- endif -%}
|
842 |
+
{%- endfor -%}
|
843 |
+
{%- if not ns.found -%}
|
844 |
+
{{- '<|PRE-SYSTEM|>' + '<|SYSTEM-MESSAGE|>' + '<|POST-SYSTEM|>' -}}
|
845 |
+
{%- endif %}
|
846 |
+
{%- for message in messages %}
|
847 |
+
{%- if message['role'] == 'system' -%}
|
848 |
+
{{- '<|PRE-SYSTEM|>' + message['content'] + '<|POST-SYSTEM|>' -}}
|
849 |
+
{%- else -%}
|
850 |
+
{%- if message['role'] == 'user' -%}
|
851 |
+
{{-'<|PRE-USER|>' + message['content'] + '<|POST-USER|>'-}}
|
852 |
+
{%- else -%}
|
853 |
+
{{-'<|PRE-ASSISTANT|>' + message['content'] + '<|POST-ASSISTANT|>' -}}
|
854 |
+
{%- endif -%}
|
855 |
+
{%- endif -%}
|
856 |
+
{%- endfor -%}
|
857 |
+
{%- if add_generation_prompt -%}
|
858 |
+
{{-'<|PRE-ASSISTANT-GENERATE|>'-}}
|
859 |
+
{%- endif -%}
|
860 |
+
"""
|
861 |
+
|
862 |
+
if 'context' in params and '<|system-message|>' in params['context']:
|
863 |
+
pre_system = params['context'].split('<|system-message|>')[0]
|
864 |
+
post_system = params['context'].split('<|system-message|>')[1]
|
865 |
+
else:
|
866 |
+
pre_system = ''
|
867 |
+
post_system = ''
|
868 |
+
|
869 |
+
pre_user = params['turn_template'].split('<|user-message|>')[0].replace('<|user|>', params['user'])
|
870 |
+
post_user = params['turn_template'].split('<|user-message|>')[1].split('<|bot|>')[0]
|
871 |
+
|
872 |
+
pre_assistant = '<|bot|>' + params['turn_template'].split('<|bot-message|>')[0].split('<|bot|>')[1]
|
873 |
+
pre_assistant = pre_assistant.replace('<|bot|>', params['bot'])
|
874 |
+
post_assistant = params['turn_template'].split('<|bot-message|>')[1]
|
875 |
+
|
876 |
+
def preprocess(string):
|
877 |
+
return string.replace('\n', '\\n').replace('\'', '\\\'')
|
878 |
+
|
879 |
+
pre_system = preprocess(pre_system)
|
880 |
+
post_system = preprocess(post_system)
|
881 |
+
pre_user = preprocess(pre_user)
|
882 |
+
post_user = preprocess(post_user)
|
883 |
+
pre_assistant = preprocess(pre_assistant)
|
884 |
+
post_assistant = preprocess(post_assistant)
|
885 |
+
|
886 |
+
if verbose:
|
887 |
+
print(
|
888 |
+
'\n',
|
889 |
+
repr(pre_system) + '\n',
|
890 |
+
repr(post_system) + '\n',
|
891 |
+
repr(pre_user) + '\n',
|
892 |
+
repr(post_user) + '\n',
|
893 |
+
repr(pre_assistant) + '\n',
|
894 |
+
repr(post_assistant) + '\n',
|
895 |
+
)
|
896 |
+
|
897 |
+
result = MASTER_TEMPLATE
|
898 |
+
if 'system_message' in params:
|
899 |
+
result = result.replace('<|SYSTEM-MESSAGE|>', preprocess(params['system_message']))
|
900 |
+
else:
|
901 |
+
result = result.replace('<|SYSTEM-MESSAGE|>', '')
|
902 |
+
|
903 |
+
result = result.replace('<|PRE-SYSTEM|>', pre_system)
|
904 |
+
result = result.replace('<|POST-SYSTEM|>', post_system)
|
905 |
+
result = result.replace('<|PRE-USER|>', pre_user)
|
906 |
+
result = result.replace('<|POST-USER|>', post_user)
|
907 |
+
result = result.replace('<|PRE-ASSISTANT|>', pre_assistant)
|
908 |
+
result = result.replace('<|PRE-ASSISTANT-GENERATE|>', pre_assistant.rstrip(' '))
|
909 |
+
result = result.replace('<|POST-ASSISTANT|>', post_assistant)
|
910 |
+
|
911 |
+
result = result.strip()
|
912 |
+
|
913 |
+
return result
|
914 |
+
|
915 |
+
|
916 |
+
def my_yaml_output(data):
|
917 |
+
'''
|
918 |
+
pyyaml is very inconsistent with multiline strings.
|
919 |
+
for simple instruction template outputs, this is enough.
|
920 |
+
'''
|
921 |
+
result = ""
|
922 |
+
for k in data:
|
923 |
+
result += k + ": |-\n"
|
924 |
+
for line in data[k].splitlines():
|
925 |
+
result += " " + line.rstrip(' ') + "\n"
|
926 |
+
|
927 |
+
return result
|
modules/ctransformers_model.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from ctransformers import AutoConfig, AutoModelForCausalLM
|
2 |
+
|
3 |
+
from modules import shared
|
4 |
+
from modules.callbacks import Iteratorize
|
5 |
+
from modules.logging_colors import logger
|
6 |
+
|
7 |
+
|
8 |
+
class CtransformersModel:
|
9 |
+
def __init__(self):
|
10 |
+
pass
|
11 |
+
|
12 |
+
@classmethod
|
13 |
+
def from_pretrained(cls, path):
|
14 |
+
result = cls()
|
15 |
+
|
16 |
+
config = AutoConfig.from_pretrained(
|
17 |
+
str(path),
|
18 |
+
threads=shared.args.threads if shared.args.threads != 0 else -1,
|
19 |
+
gpu_layers=shared.args.n_gpu_layers,
|
20 |
+
batch_size=shared.args.n_batch,
|
21 |
+
context_length=shared.args.n_ctx,
|
22 |
+
stream=True,
|
23 |
+
mmap=not shared.args.no_mmap,
|
24 |
+
mlock=shared.args.mlock
|
25 |
+
)
|
26 |
+
|
27 |
+
result.model = AutoModelForCausalLM.from_pretrained(
|
28 |
+
str(result.model_dir(path) if result.model_type_is_auto() else path),
|
29 |
+
model_type=(None if result.model_type_is_auto() else shared.args.model_type),
|
30 |
+
config=config
|
31 |
+
)
|
32 |
+
|
33 |
+
logger.info(f'Using ctransformers model_type: {result.model.model_type} for {result.model.model_path}')
|
34 |
+
return result, result
|
35 |
+
|
36 |
+
def model_type_is_auto(self):
|
37 |
+
return shared.args.model_type is None or shared.args.model_type == "Auto" or shared.args.model_type == "None"
|
38 |
+
|
39 |
+
def model_dir(self, path):
|
40 |
+
if path.is_file():
|
41 |
+
return path.parent
|
42 |
+
|
43 |
+
return path
|
44 |
+
|
45 |
+
def encode(self, string, **kwargs):
|
46 |
+
return self.model.tokenize(string)
|
47 |
+
|
48 |
+
def decode(self, ids):
|
49 |
+
return self.model.detokenize(ids)
|
50 |
+
|
51 |
+
def generate(self, prompt, state, callback=None):
|
52 |
+
prompt = prompt if type(prompt) is str else prompt.decode()
|
53 |
+
# ctransformers uses -1 for random seed
|
54 |
+
generator = self.model(
|
55 |
+
prompt=prompt,
|
56 |
+
max_new_tokens=state['max_new_tokens'],
|
57 |
+
temperature=state['temperature'],
|
58 |
+
top_p=state['top_p'],
|
59 |
+
top_k=state['top_k'],
|
60 |
+
repetition_penalty=state['repetition_penalty'],
|
61 |
+
last_n_tokens=state['repetition_penalty_range'],
|
62 |
+
seed=int(state['seed'])
|
63 |
+
)
|
64 |
+
|
65 |
+
output = ""
|
66 |
+
for token in generator:
|
67 |
+
if callback:
|
68 |
+
callback(token)
|
69 |
+
|
70 |
+
output += token
|
71 |
+
|
72 |
+
return output
|
73 |
+
|
74 |
+
def generate_with_streaming(self, *args, **kwargs):
|
75 |
+
with Iteratorize(self.generate, args, kwargs, callback=None) as generator:
|
76 |
+
reply = ''
|
77 |
+
for token in generator:
|
78 |
+
reply += token
|
79 |
+
yield reply
|
modules/deepspeed_parameters.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
def generate_ds_config(ds_bf16, train_batch_size, nvme_offload_dir):
|
2 |
+
'''
|
3 |
+
DeepSpeed configuration
|
4 |
+
https://huggingface.co/docs/transformers/main_classes/deepspeed
|
5 |
+
'''
|
6 |
+
|
7 |
+
if nvme_offload_dir:
|
8 |
+
ds_config = {
|
9 |
+
"fp16": {
|
10 |
+
"enabled": not ds_bf16,
|
11 |
+
},
|
12 |
+
"bf16": {
|
13 |
+
"enabled": ds_bf16,
|
14 |
+
},
|
15 |
+
"zero_optimization": {
|
16 |
+
"stage": 3,
|
17 |
+
"offload_param": {
|
18 |
+
"device": "nvme",
|
19 |
+
"nvme_path": nvme_offload_dir,
|
20 |
+
"pin_memory": True,
|
21 |
+
"buffer_count": 5,
|
22 |
+
"buffer_size": 1e9,
|
23 |
+
"max_in_cpu": 1e9
|
24 |
+
},
|
25 |
+
"overlap_comm": True,
|
26 |
+
"reduce_bucket_size": "auto",
|
27 |
+
"contiguous_gradients": True,
|
28 |
+
"sub_group_size": 1e8,
|
29 |
+
"stage3_prefetch_bucket_size": "auto",
|
30 |
+
"stage3_param_persistence_threshold": "auto",
|
31 |
+
"stage3_max_live_parameters": "auto",
|
32 |
+
"stage3_max_reuse_distance": "auto",
|
33 |
+
},
|
34 |
+
"aio": {
|
35 |
+
"block_size": 262144,
|
36 |
+
"queue_depth": 32,
|
37 |
+
"thread_count": 1,
|
38 |
+
"single_submit": False,
|
39 |
+
"overlap_events": True
|
40 |
+
},
|
41 |
+
"steps_per_print": 2000,
|
42 |
+
"train_batch_size": train_batch_size,
|
43 |
+
"train_micro_batch_size_per_gpu": 1,
|
44 |
+
"wall_clock_breakdown": False
|
45 |
+
}
|
46 |
+
else:
|
47 |
+
ds_config = {
|
48 |
+
"fp16": {
|
49 |
+
"enabled": not ds_bf16,
|
50 |
+
},
|
51 |
+
"bf16": {
|
52 |
+
"enabled": ds_bf16,
|
53 |
+
},
|
54 |
+
"zero_optimization": {
|
55 |
+
"stage": 3,
|
56 |
+
"offload_param": {
|
57 |
+
"device": "cpu",
|
58 |
+
"pin_memory": True
|
59 |
+
},
|
60 |
+
"overlap_comm": True,
|
61 |
+
"contiguous_gradients": True,
|
62 |
+
"reduce_bucket_size": "auto",
|
63 |
+
"stage3_prefetch_bucket_size": "auto",
|
64 |
+
"stage3_param_persistence_threshold": "auto",
|
65 |
+
"stage3_max_live_parameters": "auto",
|
66 |
+
"stage3_max_reuse_distance": "auto",
|
67 |
+
},
|
68 |
+
"steps_per_print": 2000,
|
69 |
+
"train_batch_size": train_batch_size,
|
70 |
+
"train_micro_batch_size_per_gpu": 1,
|
71 |
+
"wall_clock_breakdown": False
|
72 |
+
}
|
73 |
+
|
74 |
+
return ds_config
|
modules/evaluate.py
ADDED
@@ -0,0 +1,153 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datetime
|
2 |
+
from pathlib import Path
|
3 |
+
|
4 |
+
import pandas as pd
|
5 |
+
import torch
|
6 |
+
from datasets import load_dataset
|
7 |
+
from tqdm import tqdm
|
8 |
+
|
9 |
+
from modules import shared
|
10 |
+
from modules.logging_colors import logger
|
11 |
+
from modules.models import clear_torch_cache, load_model, unload_model
|
12 |
+
from modules.models_settings import get_model_metadata, update_model_parameters
|
13 |
+
from modules.text_generation import encode
|
14 |
+
|
15 |
+
|
16 |
+
def load_past_evaluations():
|
17 |
+
if Path('logs/evaluations.csv').exists():
|
18 |
+
df = pd.read_csv(Path('logs/evaluations.csv'), dtype=str)
|
19 |
+
df['Perplexity'] = pd.to_numeric(df['Perplexity'])
|
20 |
+
return df
|
21 |
+
else:
|
22 |
+
return pd.DataFrame(columns=['Model', 'LoRAs', 'Dataset', 'Perplexity', 'stride', 'max_length', 'Date', 'Comment'])
|
23 |
+
|
24 |
+
|
25 |
+
past_evaluations = load_past_evaluations()
|
26 |
+
|
27 |
+
|
28 |
+
def save_past_evaluations(df):
|
29 |
+
global past_evaluations
|
30 |
+
past_evaluations = df
|
31 |
+
filepath = Path('logs/evaluations.csv')
|
32 |
+
filepath.parent.mkdir(parents=True, exist_ok=True)
|
33 |
+
df.to_csv(filepath, index=False)
|
34 |
+
|
35 |
+
|
36 |
+
def calculate_perplexity(models, input_dataset, stride, _max_length):
|
37 |
+
'''
|
38 |
+
Based on:
|
39 |
+
https://huggingface.co/docs/transformers/perplexity#calculating-ppl-with-fixedlength-models
|
40 |
+
'''
|
41 |
+
|
42 |
+
if not shared.args.no_use_fast:
|
43 |
+
logger.warning("--no_use_fast is not being used. If tokenizing the input dataset takes a long time, consider loading the model with that option checked.")
|
44 |
+
|
45 |
+
global past_evaluations
|
46 |
+
cumulative_log = ''
|
47 |
+
cumulative_log += "Loading the input dataset...\n\n"
|
48 |
+
yield cumulative_log
|
49 |
+
|
50 |
+
# Copied from https://github.com/qwopqwop200/GPTQ-for-LLaMa/blob/triton/utils/datautils.py
|
51 |
+
if input_dataset == 'wikitext':
|
52 |
+
data = load_dataset('wikitext', 'wikitext-2-raw-v1', split='test')
|
53 |
+
text = "\n\n".join(data['text'])
|
54 |
+
elif input_dataset == 'ptb':
|
55 |
+
data = load_dataset('ptb_text_only', 'penn_treebank', split='validation')
|
56 |
+
text = "\n\n".join(data['sentence'])
|
57 |
+
elif input_dataset == 'ptb_new':
|
58 |
+
data = load_dataset('ptb_text_only', 'penn_treebank', split='test')
|
59 |
+
text = " ".join(data['sentence'])
|
60 |
+
else:
|
61 |
+
with open(Path(f'training/datasets/{input_dataset}.txt'), 'r', encoding='utf-8') as f:
|
62 |
+
text = f.read()
|
63 |
+
|
64 |
+
for model in models:
|
65 |
+
if is_in_past_evaluations(model, input_dataset, stride, _max_length):
|
66 |
+
cumulative_log += f"`{model}` has already been tested. Ignoring.\n\n"
|
67 |
+
yield cumulative_log
|
68 |
+
continue
|
69 |
+
|
70 |
+
if model != 'current model':
|
71 |
+
try:
|
72 |
+
yield cumulative_log + f"Loading `{model}`...\n\n"
|
73 |
+
model_settings = get_model_metadata(model)
|
74 |
+
shared.settings.update({k: v for k, v in model_settings.items() if k in shared.settings}) # hijacking the interface defaults
|
75 |
+
update_model_parameters(model_settings) # hijacking the command-line arguments
|
76 |
+
unload_model()
|
77 |
+
shared.model, shared.tokenizer = load_model(model)
|
78 |
+
except:
|
79 |
+
cumulative_log += f"Failed to load `{model}`. Moving on.\n\n"
|
80 |
+
yield cumulative_log
|
81 |
+
continue
|
82 |
+
|
83 |
+
cumulative_log += f"Processing `{shared.model_name}`...\n\n"
|
84 |
+
yield cumulative_log + "Tokenizing the input dataset...\n\n"
|
85 |
+
encodings = encode(text, add_special_tokens=False)
|
86 |
+
seq_len = encodings.shape[1]
|
87 |
+
if _max_length:
|
88 |
+
max_length = _max_length
|
89 |
+
elif hasattr(shared.model.config, 'max_position_embeddings'):
|
90 |
+
max_length = shared.model.config.max_position_embeddings
|
91 |
+
else:
|
92 |
+
max_length = 2048
|
93 |
+
|
94 |
+
nlls = []
|
95 |
+
prev_end_loc = 0
|
96 |
+
for begin_loc in tqdm(range(0, seq_len, stride)):
|
97 |
+
yield cumulative_log + f"Evaluating... {100*begin_loc/seq_len:.2f}%"
|
98 |
+
end_loc = min(begin_loc + max_length, seq_len)
|
99 |
+
trg_len = end_loc - prev_end_loc # may be different from stride on last loop
|
100 |
+
input_ids = encodings[:, begin_loc:end_loc]
|
101 |
+
target_ids = input_ids.clone()
|
102 |
+
target_ids[:, :-trg_len] = -100
|
103 |
+
clear_torch_cache()
|
104 |
+
with torch.no_grad():
|
105 |
+
outputs = shared.model(input_ids=input_ids, labels=target_ids)
|
106 |
+
|
107 |
+
# loss is calculated using CrossEntropyLoss which averages over valid labels
|
108 |
+
# N.B. the model only calculates loss over trg_len - 1 labels, because it internally shifts the labels
|
109 |
+
# to the left by 1.
|
110 |
+
neg_log_likelihood = outputs.loss
|
111 |
+
|
112 |
+
nlls.append(neg_log_likelihood)
|
113 |
+
prev_end_loc = end_loc
|
114 |
+
if end_loc == seq_len:
|
115 |
+
break
|
116 |
+
|
117 |
+
ppl = torch.exp(torch.stack(nlls).mean())
|
118 |
+
add_entry_to_past_evaluations(float(ppl), shared.model_name, input_dataset, stride, _max_length)
|
119 |
+
save_past_evaluations(past_evaluations)
|
120 |
+
cumulative_log += f"The perplexity for `{shared.model_name}` is: {float(ppl)}\n\n"
|
121 |
+
yield cumulative_log
|
122 |
+
|
123 |
+
|
124 |
+
def add_entry_to_past_evaluations(perplexity, model, dataset, stride, max_length):
|
125 |
+
global past_evaluations
|
126 |
+
entry = {
|
127 |
+
'Model': model,
|
128 |
+
'LoRAs': ', '.join(shared.lora_names) or '-',
|
129 |
+
'Dataset': dataset,
|
130 |
+
'Perplexity': perplexity,
|
131 |
+
'stride': str(stride),
|
132 |
+
'max_length': str(max_length),
|
133 |
+
'Date': datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'),
|
134 |
+
'Comment': ''
|
135 |
+
}
|
136 |
+
past_evaluations = pd.concat([past_evaluations, pd.DataFrame([entry])], ignore_index=True)
|
137 |
+
|
138 |
+
|
139 |
+
def is_in_past_evaluations(model, dataset, stride, max_length):
|
140 |
+
entries = past_evaluations[(past_evaluations['Model'] == model) &
|
141 |
+
(past_evaluations['Dataset'] == dataset) &
|
142 |
+
(past_evaluations['max_length'] == str(max_length)) &
|
143 |
+
(past_evaluations['stride'] == str(stride))]
|
144 |
+
|
145 |
+
if entries.shape[0] > 0:
|
146 |
+
return True
|
147 |
+
else:
|
148 |
+
return False
|
149 |
+
|
150 |
+
|
151 |
+
def generate_markdown_table():
|
152 |
+
sorted_df = past_evaluations.sort_values(by=['Dataset', 'stride', 'Perplexity', 'Date'])
|
153 |
+
return sorted_df
|
modules/exllamav2.py
ADDED
@@ -0,0 +1,149 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import traceback
|
2 |
+
from pathlib import Path
|
3 |
+
|
4 |
+
import torch
|
5 |
+
from exllamav2 import (
|
6 |
+
ExLlamaV2,
|
7 |
+
ExLlamaV2Cache,
|
8 |
+
ExLlamaV2Cache_8bit,
|
9 |
+
ExLlamaV2Config,
|
10 |
+
ExLlamaV2Tokenizer
|
11 |
+
)
|
12 |
+
from exllamav2.generator import ExLlamaV2Sampler, ExLlamaV2StreamingGenerator
|
13 |
+
|
14 |
+
from modules import shared
|
15 |
+
from modules.logging_colors import logger
|
16 |
+
from modules.text_generation import get_max_prompt_length
|
17 |
+
|
18 |
+
try:
|
19 |
+
import flash_attn
|
20 |
+
except ModuleNotFoundError:
|
21 |
+
logger.warning(
|
22 |
+
'You are running ExLlamaV2 without flash-attention. This will cause the VRAM usage '
|
23 |
+
'to be a lot higher than it could be.\n'
|
24 |
+
'Try installing flash-attention following the instructions here: '
|
25 |
+
'https://github.com/Dao-AILab/flash-attention#installation-and-features'
|
26 |
+
)
|
27 |
+
pass
|
28 |
+
except Exception:
|
29 |
+
logger.warning('Failed to load flash-attention due to the following error:\n')
|
30 |
+
traceback.print_exc()
|
31 |
+
|
32 |
+
|
33 |
+
class Exllamav2Model:
|
34 |
+
def __init__(self):
|
35 |
+
pass
|
36 |
+
|
37 |
+
@classmethod
|
38 |
+
def from_pretrained(self, path_to_model):
|
39 |
+
|
40 |
+
path_to_model = Path(f'{shared.args.model_dir}') / Path(path_to_model)
|
41 |
+
|
42 |
+
config = ExLlamaV2Config()
|
43 |
+
config.model_dir = str(path_to_model)
|
44 |
+
config.prepare()
|
45 |
+
|
46 |
+
config.max_seq_len = shared.args.max_seq_len
|
47 |
+
config.scale_pos_emb = shared.args.compress_pos_emb
|
48 |
+
config.scale_alpha_value = shared.args.alpha_value
|
49 |
+
config.no_flash_attn = shared.args.no_flash_attn
|
50 |
+
config.num_experts_per_token = int(shared.args.num_experts_per_token)
|
51 |
+
|
52 |
+
model = ExLlamaV2(config)
|
53 |
+
|
54 |
+
split = None
|
55 |
+
if shared.args.gpu_split:
|
56 |
+
split = [float(alloc) for alloc in shared.args.gpu_split.split(",")]
|
57 |
+
|
58 |
+
model.load(split)
|
59 |
+
|
60 |
+
tokenizer = ExLlamaV2Tokenizer(config)
|
61 |
+
if shared.args.cache_8bit:
|
62 |
+
cache = ExLlamaV2Cache_8bit(model)
|
63 |
+
else:
|
64 |
+
cache = ExLlamaV2Cache(model)
|
65 |
+
|
66 |
+
generator = ExLlamaV2StreamingGenerator(model, cache, tokenizer)
|
67 |
+
|
68 |
+
result = self()
|
69 |
+
result.model = model
|
70 |
+
result.cache = cache
|
71 |
+
result.tokenizer = tokenizer
|
72 |
+
result.generator = generator
|
73 |
+
result.loras = None
|
74 |
+
return result, result
|
75 |
+
|
76 |
+
def encode(self, string, **kwargs):
|
77 |
+
return self.tokenizer.encode(string, add_bos=True, encode_special_tokens=True)
|
78 |
+
|
79 |
+
def decode(self, ids, **kwargs):
|
80 |
+
if isinstance(ids, list):
|
81 |
+
ids = torch.tensor([ids])
|
82 |
+
elif isinstance(ids, torch.Tensor) and ids.numel() == 1:
|
83 |
+
ids = ids.view(1, -1)
|
84 |
+
|
85 |
+
return self.tokenizer.decode(ids, decode_special_tokens=True)[0]
|
86 |
+
|
87 |
+
def get_logits(self, token_ids, **kwargs):
|
88 |
+
self.cache.current_seq_len = 0
|
89 |
+
if token_ids.shape[-1] > 1:
|
90 |
+
self.model.forward(token_ids[:, :-1], self.cache, input_mask=None, preprocess_only=True, loras=self.loras)
|
91 |
+
|
92 |
+
return self.model.forward(token_ids[:, -1:], self.cache, input_mask=None, loras=self.loras, **kwargs).float().cpu()
|
93 |
+
|
94 |
+
def generate_with_streaming(self, prompt, state):
|
95 |
+
settings = ExLlamaV2Sampler.Settings()
|
96 |
+
|
97 |
+
settings.token_repetition_penalty = state['repetition_penalty']
|
98 |
+
settings.token_repetition_range = -1 if state['repetition_penalty_range'] <= 0 else state['repetition_penalty_range']
|
99 |
+
|
100 |
+
settings.token_frequency_penalty = state['frequency_penalty']
|
101 |
+
settings.token_presence_penalty = state['presence_penalty']
|
102 |
+
|
103 |
+
settings.temperature = state['temperature']
|
104 |
+
settings.top_k = state['top_k']
|
105 |
+
settings.top_p = state['top_p']
|
106 |
+
settings.top_a = state['top_a']
|
107 |
+
settings.min_p = state['min_p']
|
108 |
+
settings.tfs = state['tfs']
|
109 |
+
settings.typical = state['typical_p']
|
110 |
+
|
111 |
+
settings.temperature_last = state['temperature_last']
|
112 |
+
|
113 |
+
settings.mirostat = state['mirostat_mode'] == 2
|
114 |
+
settings.mirostat_tau = state['mirostat_tau']
|
115 |
+
settings.mirostat_eta = state['mirostat_eta']
|
116 |
+
|
117 |
+
if state['ban_eos_token']:
|
118 |
+
settings.disallow_tokens(self.tokenizer, [self.tokenizer.eos_token_id])
|
119 |
+
|
120 |
+
if state['custom_token_bans']:
|
121 |
+
to_ban = [int(x) for x in state['custom_token_bans'].split(',')]
|
122 |
+
if len(to_ban) > 0:
|
123 |
+
settings.disallow_tokens(self.tokenizer, to_ban)
|
124 |
+
|
125 |
+
ids = self.tokenizer.encode(prompt, add_bos=state['add_bos_token'], encode_special_tokens=True)
|
126 |
+
ids = ids[:, -get_max_prompt_length(state):]
|
127 |
+
|
128 |
+
if state['auto_max_new_tokens']:
|
129 |
+
max_new_tokens = state['truncation_length'] - ids.shape[-1]
|
130 |
+
else:
|
131 |
+
max_new_tokens = state['max_new_tokens']
|
132 |
+
|
133 |
+
self.generator.begin_stream(ids, settings, loras=self.loras)
|
134 |
+
|
135 |
+
decoded_text = ''
|
136 |
+
for i in range(max_new_tokens):
|
137 |
+
chunk, eos, _ = self.generator.stream()
|
138 |
+
if eos or shared.stop_everything:
|
139 |
+
break
|
140 |
+
|
141 |
+
decoded_text += chunk
|
142 |
+
yield decoded_text
|
143 |
+
|
144 |
+
def generate(self, prompt, state):
|
145 |
+
output = ''
|
146 |
+
for output in self.generate_with_streaming(prompt, state):
|
147 |
+
pass
|
148 |
+
|
149 |
+
return output
|
modules/exllamav2_hf.py
ADDED
@@ -0,0 +1,170 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import traceback
|
3 |
+
from pathlib import Path
|
4 |
+
from typing import Any, Dict, Optional, Union
|
5 |
+
|
6 |
+
import torch
|
7 |
+
from exllamav2 import (
|
8 |
+
ExLlamaV2,
|
9 |
+
ExLlamaV2Cache,
|
10 |
+
ExLlamaV2Cache_8bit,
|
11 |
+
ExLlamaV2Config
|
12 |
+
)
|
13 |
+
from torch.nn import CrossEntropyLoss
|
14 |
+
from transformers import GenerationConfig, PretrainedConfig, PreTrainedModel
|
15 |
+
from transformers.modeling_outputs import CausalLMOutputWithPast
|
16 |
+
|
17 |
+
from modules import shared
|
18 |
+
from modules.logging_colors import logger
|
19 |
+
|
20 |
+
try:
|
21 |
+
import flash_attn
|
22 |
+
except ModuleNotFoundError:
|
23 |
+
logger.warning(
|
24 |
+
'You are running ExLlamaV2 without flash-attention. This will cause the VRAM usage '
|
25 |
+
'to be a lot higher than it could be.\n'
|
26 |
+
'Try installing flash-attention following the instructions here: '
|
27 |
+
'https://github.com/Dao-AILab/flash-attention#installation-and-features'
|
28 |
+
)
|
29 |
+
pass
|
30 |
+
except Exception:
|
31 |
+
logger.warning('Failed to load flash-attention due to the following error:\n')
|
32 |
+
traceback.print_exc()
|
33 |
+
|
34 |
+
|
35 |
+
class Exllamav2HF(PreTrainedModel):
|
36 |
+
def __init__(self, config: ExLlamaV2Config):
|
37 |
+
super().__init__(PretrainedConfig())
|
38 |
+
self.ex_config = config
|
39 |
+
self.ex_model = ExLlamaV2(config)
|
40 |
+
split = None
|
41 |
+
if shared.args.gpu_split:
|
42 |
+
split = [float(alloc) for alloc in shared.args.gpu_split.split(",")]
|
43 |
+
|
44 |
+
self.ex_model.load(split)
|
45 |
+
self.generation_config = GenerationConfig()
|
46 |
+
self.loras = None
|
47 |
+
|
48 |
+
if shared.args.cache_8bit:
|
49 |
+
self.ex_cache = ExLlamaV2Cache_8bit(self.ex_model)
|
50 |
+
else:
|
51 |
+
self.ex_cache = ExLlamaV2Cache(self.ex_model)
|
52 |
+
|
53 |
+
self.past_seq = None
|
54 |
+
if shared.args.cfg_cache:
|
55 |
+
if shared.args.cache_8bit:
|
56 |
+
self.ex_cache_negative = ExLlamaV2Cache_8bit(self.ex_model)
|
57 |
+
else:
|
58 |
+
self.ex_cache_negative = ExLlamaV2Cache(self.ex_model)
|
59 |
+
|
60 |
+
self.past_seq_negative = None
|
61 |
+
|
62 |
+
def _validate_model_class(self):
|
63 |
+
pass
|
64 |
+
|
65 |
+
def _validate_model_kwargs(self, model_kwargs: Dict[str, Any]):
|
66 |
+
pass
|
67 |
+
|
68 |
+
def prepare_inputs_for_generation(self, input_ids, **kwargs):
|
69 |
+
return {'input_ids': input_ids, **kwargs}
|
70 |
+
|
71 |
+
@property
|
72 |
+
def device(self) -> torch.device:
|
73 |
+
return torch.device(0)
|
74 |
+
|
75 |
+
def __call__(self, *args, **kwargs):
|
76 |
+
use_cache = kwargs.get('use_cache', True)
|
77 |
+
labels = kwargs.get('labels', None)
|
78 |
+
past_key_values = kwargs.get('past_key_values', None)
|
79 |
+
|
80 |
+
if len(args) > 0:
|
81 |
+
if not shared.args.cfg_cache:
|
82 |
+
logger.error("Please enable the cfg-cache option to use CFG with ExLlamav2_HF.")
|
83 |
+
return
|
84 |
+
|
85 |
+
input_ids = args[0]
|
86 |
+
is_negative = True
|
87 |
+
past_seq = self.past_seq_negative
|
88 |
+
ex_cache = self.ex_cache_negative
|
89 |
+
else:
|
90 |
+
input_ids = kwargs['input_ids']
|
91 |
+
is_negative = False
|
92 |
+
past_seq = self.past_seq
|
93 |
+
ex_cache = self.ex_cache
|
94 |
+
|
95 |
+
seq = input_ids[0].tolist()
|
96 |
+
if is_negative and past_key_values is not None:
|
97 |
+
seq = past_key_values + seq
|
98 |
+
|
99 |
+
seq_tensor = torch.tensor(seq)
|
100 |
+
reset = True
|
101 |
+
|
102 |
+
# Make the forward call
|
103 |
+
if labels is None:
|
104 |
+
if past_seq is not None:
|
105 |
+
min_length = min(past_seq.shape[0], seq_tensor.shape[0])
|
106 |
+
indices = torch.nonzero(~torch.eq(past_seq[:min_length], seq_tensor[:min_length]))
|
107 |
+
if len(indices) > 0:
|
108 |
+
longest_prefix = indices[0].item()
|
109 |
+
else:
|
110 |
+
longest_prefix = min_length
|
111 |
+
|
112 |
+
if longest_prefix > 0:
|
113 |
+
reset = False
|
114 |
+
ex_cache.current_seq_len = longest_prefix
|
115 |
+
if len(seq_tensor) - longest_prefix > 1:
|
116 |
+
self.ex_model.forward(seq_tensor[longest_prefix:-1].view(1, -1), ex_cache, preprocess_only=True, loras=self.loras)
|
117 |
+
elif len(seq_tensor) == longest_prefix:
|
118 |
+
# Very tricky: if the prefix we are reusing *is* the input_ids, then we have to back up the cache pointer by one,
|
119 |
+
# because we feed input_ids[-1] to forward() below, but that last token is already in the cache!
|
120 |
+
ex_cache.current_seq_len -= 1
|
121 |
+
|
122 |
+
if reset:
|
123 |
+
ex_cache.current_seq_len = 0
|
124 |
+
if len(seq_tensor) > 1:
|
125 |
+
self.ex_model.forward(seq_tensor[:-1].view(1, -1), ex_cache, preprocess_only=True, loras=self.loras)
|
126 |
+
|
127 |
+
logits = self.ex_model.forward(seq_tensor[-1:].view(1, -1), ex_cache, loras=self.loras).to(input_ids.device).float()
|
128 |
+
else:
|
129 |
+
ex_cache.current_seq_len = 0
|
130 |
+
logits = self.ex_model.forward(seq_tensor.view(1, -1), ex_cache, last_id_only=False, loras=self.loras).float()
|
131 |
+
|
132 |
+
if is_negative:
|
133 |
+
self.past_seq_negative = seq_tensor
|
134 |
+
else:
|
135 |
+
self.past_seq = seq_tensor
|
136 |
+
|
137 |
+
loss = None
|
138 |
+
if labels is not None:
|
139 |
+
# Shift so that tokens < n predict n
|
140 |
+
shift_logits = logits[..., :-1, :].contiguous()
|
141 |
+
shift_labels = labels[..., 1:].contiguous()
|
142 |
+
# Flatten the tokens
|
143 |
+
loss_fct = CrossEntropyLoss()
|
144 |
+
shift_logits = shift_logits.view(-1, logits.shape[-1])
|
145 |
+
shift_labels = shift_labels.view(-1)
|
146 |
+
# Enable model parallelism
|
147 |
+
shift_labels = shift_labels.to(shift_logits.device)
|
148 |
+
loss = loss_fct(shift_logits, shift_labels)
|
149 |
+
|
150 |
+
return CausalLMOutputWithPast(logits=logits, past_key_values=seq if use_cache else None, loss=loss)
|
151 |
+
|
152 |
+
@classmethod
|
153 |
+
def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.PathLike]], *model_args, **kwargs):
|
154 |
+
assert len(model_args) == 0 and len(kwargs) == 0, "extra args is currently not supported"
|
155 |
+
if isinstance(pretrained_model_name_or_path, str):
|
156 |
+
pretrained_model_name_or_path = Path(pretrained_model_name_or_path)
|
157 |
+
|
158 |
+
pretrained_model_name_or_path = Path(f'{shared.args.model_dir}') / Path(pretrained_model_name_or_path)
|
159 |
+
|
160 |
+
config = ExLlamaV2Config()
|
161 |
+
config.model_dir = str(pretrained_model_name_or_path)
|
162 |
+
config.prepare()
|
163 |
+
|
164 |
+
config.max_seq_len = shared.args.max_seq_len
|
165 |
+
config.scale_pos_emb = shared.args.compress_pos_emb
|
166 |
+
config.scale_alpha_value = shared.args.alpha_value
|
167 |
+
config.no_flash_attn = shared.args.no_flash_attn
|
168 |
+
config.num_experts_per_token = int(shared.args.num_experts_per_token)
|
169 |
+
|
170 |
+
return Exllamav2HF(config)
|
modules/extensions.py
ADDED
@@ -0,0 +1,232 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import traceback
|
2 |
+
from functools import partial
|
3 |
+
from inspect import signature
|
4 |
+
|
5 |
+
import gradio as gr
|
6 |
+
|
7 |
+
import extensions
|
8 |
+
import modules.shared as shared
|
9 |
+
from modules.logging_colors import logger
|
10 |
+
|
11 |
+
state = {}
|
12 |
+
available_extensions = []
|
13 |
+
setup_called = set()
|
14 |
+
|
15 |
+
|
16 |
+
def apply_settings(extension, name):
|
17 |
+
if not hasattr(extension, 'params'):
|
18 |
+
return
|
19 |
+
|
20 |
+
for param in extension.params:
|
21 |
+
_id = f"{name}-{param}"
|
22 |
+
shared.default_settings[_id] = extension.params[param]
|
23 |
+
if _id in shared.settings:
|
24 |
+
extension.params[param] = shared.settings[_id]
|
25 |
+
|
26 |
+
|
27 |
+
def load_extensions():
|
28 |
+
global state, setup_called
|
29 |
+
state = {}
|
30 |
+
for i, name in enumerate(shared.args.extensions):
|
31 |
+
if name in available_extensions:
|
32 |
+
if name != 'api':
|
33 |
+
logger.info(f'Loading the extension "{name}"')
|
34 |
+
try:
|
35 |
+
try:
|
36 |
+
exec(f"import extensions.{name}.script")
|
37 |
+
except ModuleNotFoundError:
|
38 |
+
logger.error(f"Could not import the requirements for '{name}'. Make sure to install the requirements for the extension.\n\nLinux / Mac:\n\npip install -r extensions/{name}/requirements.txt --upgrade\n\nWindows:\n\npip install -r extensions\\{name}\\requirements.txt --upgrade\n\nIf you used the one-click installer, paste the command above in the terminal window opened after launching the cmd script for your OS.")
|
39 |
+
raise
|
40 |
+
|
41 |
+
extension = getattr(extensions, name).script
|
42 |
+
|
43 |
+
# Only run setup() and apply settings from settings.yaml once
|
44 |
+
if extension not in setup_called:
|
45 |
+
apply_settings(extension, name)
|
46 |
+
if hasattr(extension, "setup"):
|
47 |
+
extension.setup()
|
48 |
+
|
49 |
+
setup_called.add(extension)
|
50 |
+
|
51 |
+
state[name] = [True, i]
|
52 |
+
except:
|
53 |
+
logger.error(f'Failed to load the extension "{name}".')
|
54 |
+
traceback.print_exc()
|
55 |
+
|
56 |
+
|
57 |
+
# This iterator returns the extensions in the order specified in the command-line
|
58 |
+
def iterator():
|
59 |
+
for name in sorted(state, key=lambda x: state[x][1]):
|
60 |
+
if state[name][0]:
|
61 |
+
yield getattr(extensions, name).script, name
|
62 |
+
|
63 |
+
|
64 |
+
# Extension functions that map string -> string
|
65 |
+
def _apply_string_extensions(function_name, text, state, is_chat=False):
|
66 |
+
for extension, _ in iterator():
|
67 |
+
if hasattr(extension, function_name):
|
68 |
+
func = getattr(extension, function_name)
|
69 |
+
|
70 |
+
# Handle old extensions without the 'state' arg or
|
71 |
+
# the 'is_chat' kwarg
|
72 |
+
count = 0
|
73 |
+
has_chat = False
|
74 |
+
for k in signature(func).parameters:
|
75 |
+
if k == 'is_chat':
|
76 |
+
has_chat = True
|
77 |
+
else:
|
78 |
+
count += 1
|
79 |
+
|
80 |
+
if count == 2:
|
81 |
+
args = [text, state]
|
82 |
+
else:
|
83 |
+
args = [text]
|
84 |
+
|
85 |
+
if has_chat:
|
86 |
+
kwargs = {'is_chat': is_chat}
|
87 |
+
else:
|
88 |
+
kwargs = {}
|
89 |
+
|
90 |
+
text = func(*args, **kwargs)
|
91 |
+
|
92 |
+
return text
|
93 |
+
|
94 |
+
|
95 |
+
# Extension functions that map string -> string
|
96 |
+
def _apply_chat_input_extensions(text, visible_text, state):
|
97 |
+
for extension, _ in iterator():
|
98 |
+
if hasattr(extension, 'chat_input_modifier'):
|
99 |
+
text, visible_text = extension.chat_input_modifier(text, visible_text, state)
|
100 |
+
|
101 |
+
return text, visible_text
|
102 |
+
|
103 |
+
|
104 |
+
# custom_generate_chat_prompt handling - currently only the first one will work
|
105 |
+
def _apply_custom_generate_chat_prompt(text, state, **kwargs):
|
106 |
+
for extension, _ in iterator():
|
107 |
+
if hasattr(extension, 'custom_generate_chat_prompt'):
|
108 |
+
return extension.custom_generate_chat_prompt(text, state, **kwargs)
|
109 |
+
|
110 |
+
return None
|
111 |
+
|
112 |
+
|
113 |
+
# Extension that modifies the input parameters before they are used
|
114 |
+
def _apply_state_modifier_extensions(state):
|
115 |
+
for extension, _ in iterator():
|
116 |
+
if hasattr(extension, "state_modifier"):
|
117 |
+
state = getattr(extension, "state_modifier")(state)
|
118 |
+
|
119 |
+
return state
|
120 |
+
|
121 |
+
|
122 |
+
# Extension that modifies the chat history before it is used
|
123 |
+
def _apply_history_modifier_extensions(history):
|
124 |
+
for extension, _ in iterator():
|
125 |
+
if hasattr(extension, "history_modifier"):
|
126 |
+
history = getattr(extension, "history_modifier")(history)
|
127 |
+
|
128 |
+
return history
|
129 |
+
|
130 |
+
|
131 |
+
# Extension functions that override the default tokenizer output - The order of execution is not defined
|
132 |
+
def _apply_tokenizer_extensions(function_name, state, prompt, input_ids, input_embeds):
|
133 |
+
for extension, _ in iterator():
|
134 |
+
if hasattr(extension, function_name):
|
135 |
+
prompt, input_ids, input_embeds = getattr(extension, function_name)(state, prompt, input_ids, input_embeds)
|
136 |
+
|
137 |
+
return prompt, input_ids, input_embeds
|
138 |
+
|
139 |
+
|
140 |
+
# Allow extensions to add their own logits processors to the stack being run.
|
141 |
+
# Each extension would call `processor_list.append({their LogitsProcessor}())`.
|
142 |
+
def _apply_logits_processor_extensions(function_name, processor_list, input_ids):
|
143 |
+
for extension, _ in iterator():
|
144 |
+
if hasattr(extension, function_name):
|
145 |
+
result = getattr(extension, function_name)(processor_list, input_ids)
|
146 |
+
if type(result) is list:
|
147 |
+
processor_list = result
|
148 |
+
|
149 |
+
return processor_list
|
150 |
+
|
151 |
+
|
152 |
+
# Get prompt length in tokens after applying extension functions which override the default tokenizer output
|
153 |
+
# currently only the first one will work
|
154 |
+
def _apply_custom_tokenized_length(prompt):
|
155 |
+
for extension, _ in iterator():
|
156 |
+
if hasattr(extension, 'custom_tokenized_length'):
|
157 |
+
return getattr(extension, 'custom_tokenized_length')(prompt)
|
158 |
+
|
159 |
+
return None
|
160 |
+
|
161 |
+
|
162 |
+
# Custom generate reply handling - currently only the first one will work
|
163 |
+
def _apply_custom_generate_reply():
|
164 |
+
for extension, _ in iterator():
|
165 |
+
if hasattr(extension, 'custom_generate_reply'):
|
166 |
+
return getattr(extension, 'custom_generate_reply')
|
167 |
+
|
168 |
+
return None
|
169 |
+
|
170 |
+
|
171 |
+
def _apply_custom_css():
|
172 |
+
all_css = ''
|
173 |
+
for extension, _ in iterator():
|
174 |
+
if hasattr(extension, 'custom_css'):
|
175 |
+
all_css += getattr(extension, 'custom_css')()
|
176 |
+
|
177 |
+
return all_css
|
178 |
+
|
179 |
+
|
180 |
+
def _apply_custom_js():
|
181 |
+
all_js = ''
|
182 |
+
for extension, _ in iterator():
|
183 |
+
if hasattr(extension, 'custom_js'):
|
184 |
+
all_js += getattr(extension, 'custom_js')()
|
185 |
+
|
186 |
+
return all_js
|
187 |
+
|
188 |
+
|
189 |
+
def create_extensions_block():
|
190 |
+
to_display = []
|
191 |
+
for extension, name in iterator():
|
192 |
+
if hasattr(extension, "ui") and not (hasattr(extension, 'params') and extension.params.get('is_tab', False)):
|
193 |
+
to_display.append((extension, name))
|
194 |
+
|
195 |
+
# Creating the extension ui elements
|
196 |
+
if len(to_display) > 0:
|
197 |
+
with gr.Column(elem_id="extensions"):
|
198 |
+
for row in to_display:
|
199 |
+
extension, _ = row
|
200 |
+
extension.ui()
|
201 |
+
|
202 |
+
|
203 |
+
def create_extensions_tabs():
|
204 |
+
for extension, name in iterator():
|
205 |
+
if hasattr(extension, "ui") and (hasattr(extension, 'params') and extension.params.get('is_tab', False)):
|
206 |
+
display_name = getattr(extension, 'params', {}).get('display_name', name)
|
207 |
+
with gr.Tab(display_name, elem_classes="extension-tab"):
|
208 |
+
extension.ui()
|
209 |
+
|
210 |
+
|
211 |
+
EXTENSION_MAP = {
|
212 |
+
"input": partial(_apply_string_extensions, "input_modifier"),
|
213 |
+
"output": partial(_apply_string_extensions, "output_modifier"),
|
214 |
+
"chat_input": _apply_chat_input_extensions,
|
215 |
+
"state": _apply_state_modifier_extensions,
|
216 |
+
"history": _apply_history_modifier_extensions,
|
217 |
+
"bot_prefix": partial(_apply_string_extensions, "bot_prefix_modifier"),
|
218 |
+
"tokenizer": partial(_apply_tokenizer_extensions, "tokenizer_modifier"),
|
219 |
+
'logits_processor': partial(_apply_logits_processor_extensions, 'logits_processor_modifier'),
|
220 |
+
"custom_generate_chat_prompt": _apply_custom_generate_chat_prompt,
|
221 |
+
"custom_generate_reply": _apply_custom_generate_reply,
|
222 |
+
"tokenized_length": _apply_custom_tokenized_length,
|
223 |
+
"css": _apply_custom_css,
|
224 |
+
"js": _apply_custom_js
|
225 |
+
}
|
226 |
+
|
227 |
+
|
228 |
+
def apply_extensions(typ, *args, **kwargs):
|
229 |
+
if typ not in EXTENSION_MAP:
|
230 |
+
raise ValueError(f"Invalid extension type {typ}")
|
231 |
+
|
232 |
+
return EXTENSION_MAP[typ](*args, **kwargs)
|
modules/github.py
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import subprocess
|
2 |
+
from pathlib import Path
|
3 |
+
|
4 |
+
new_extensions = set()
|
5 |
+
|
6 |
+
|
7 |
+
def clone_or_pull_repository(github_url):
|
8 |
+
global new_extensions
|
9 |
+
|
10 |
+
repository_folder = Path("extensions")
|
11 |
+
repo_name = github_url.rstrip("/").split("/")[-1].split(".")[0]
|
12 |
+
|
13 |
+
# Check if the repository folder exists
|
14 |
+
if not repository_folder.exists():
|
15 |
+
repository_folder.mkdir(parents=True)
|
16 |
+
|
17 |
+
repo_path = repository_folder / repo_name
|
18 |
+
|
19 |
+
# Check if the repository is already cloned
|
20 |
+
if repo_path.exists():
|
21 |
+
yield f"Updating {github_url}..."
|
22 |
+
# Perform a 'git pull' to update the repository
|
23 |
+
try:
|
24 |
+
pull_output = subprocess.check_output(["git", "-C", repo_path, "pull"], stderr=subprocess.STDOUT)
|
25 |
+
yield "Done."
|
26 |
+
return pull_output.decode()
|
27 |
+
except subprocess.CalledProcessError as e:
|
28 |
+
return str(e)
|
29 |
+
|
30 |
+
# Clone the repository
|
31 |
+
try:
|
32 |
+
yield f"Cloning {github_url}..."
|
33 |
+
clone_output = subprocess.check_output(["git", "clone", github_url, repo_path], stderr=subprocess.STDOUT)
|
34 |
+
new_extensions.add(repo_name)
|
35 |
+
yield f"The extension `{repo_name}` has been downloaded.\n\nPlease close the the web UI completely and launch it again to be able to load it."
|
36 |
+
return clone_output.decode()
|
37 |
+
except subprocess.CalledProcessError as e:
|
38 |
+
return str(e)
|
modules/grammar/__pycache__/grammar_utils.cpython-311.pyc
ADDED
Binary file (33 kB). View file
|
|
modules/grammar/__pycache__/logits_process.cpython-311.pyc
ADDED
Binary file (5.46 kB). View file
|
|