Upload folder using huggingface_hub
Browse files- README.md +22 -1
- adapter_config.json +26 -0
- adapter_model.bin +3 -0
- xtuner_config.py +212 -0
README.md
CHANGED
@@ -1,3 +1,24 @@
|
|
1 |
---
|
2 |
-
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
library_name: peft
|
3 |
---
|
4 |
+
## Training procedure
|
5 |
+
|
6 |
+
|
7 |
+
The following `bitsandbytes` quantization config was used during training:
|
8 |
+
- quant_method: bitsandbytes
|
9 |
+
- _load_in_8bit: False
|
10 |
+
- _load_in_4bit: True
|
11 |
+
- llm_int8_threshold: 6.0
|
12 |
+
- llm_int8_skip_modules: None
|
13 |
+
- llm_int8_enable_fp32_cpu_offload: False
|
14 |
+
- llm_int8_has_fp16_weight: False
|
15 |
+
- bnb_4bit_quant_type: nf4
|
16 |
+
- bnb_4bit_use_double_quant: True
|
17 |
+
- bnb_4bit_compute_dtype: float16
|
18 |
+
- bnb_4bit_quant_storage: uint8
|
19 |
+
- load_in_4bit: True
|
20 |
+
- load_in_8bit: False
|
21 |
+
### Framework versions
|
22 |
+
|
23 |
+
|
24 |
+
- PEFT 0.5.0
|
adapter_config.json
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"auto_mapping": null,
|
3 |
+
"base_model_name_or_path": "/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473",
|
4 |
+
"bias": "none",
|
5 |
+
"fan_in_fan_out": false,
|
6 |
+
"inference_mode": true,
|
7 |
+
"init_lora_weights": true,
|
8 |
+
"layers_pattern": null,
|
9 |
+
"layers_to_transform": null,
|
10 |
+
"lora_alpha": 16,
|
11 |
+
"lora_dropout": 0.1,
|
12 |
+
"modules_to_save": null,
|
13 |
+
"peft_type": "LORA",
|
14 |
+
"r": 64,
|
15 |
+
"revision": null,
|
16 |
+
"target_modules": [
|
17 |
+
"v_proj",
|
18 |
+
"down_proj",
|
19 |
+
"up_proj",
|
20 |
+
"q_proj",
|
21 |
+
"k_proj",
|
22 |
+
"gate_proj",
|
23 |
+
"o_proj"
|
24 |
+
],
|
25 |
+
"task_type": "CAUSAL_LM"
|
26 |
+
}
|
adapter_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:0a0283577583ffcf60fe4e991f13ec0a0d15122a4b1197e45447b707392c2e8c
|
3 |
+
size 335706314
|
xtuner_config.py
ADDED
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
SYSTEM = 'Você é assistente de IA chamado ZephyrBode.\n - O ZephyrBode é um modelo de língua conversacional projetado para ser prestativo, honesto e inofensivo.\n - O ZephyrBode pode entender e se comunicar fluentemente na linguagem escolhida pelo usuário, em especial o Português e o Inglês'
|
2 |
+
accumulative_counts = 16
|
3 |
+
alpaca_en = dict(
|
4 |
+
dataset=dict(
|
5 |
+
data_files=
|
6 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/datasets--recogna-nlp--ultra-alpaca-ptbr/snapshots/e69900d074177d370a911096fc30bdf407eff666/train.json',
|
7 |
+
path='json',
|
8 |
+
type='datasets.load_dataset'),
|
9 |
+
dataset_map_fn='xtuner.dataset.map_fns.ultracabrita_map_fn',
|
10 |
+
max_length=2048,
|
11 |
+
pack_to_max_length=True,
|
12 |
+
remove_unused_columns=True,
|
13 |
+
shuffle_before_pack=True,
|
14 |
+
template_map_fn=dict(
|
15 |
+
template='xtuner.utils.PROMPT_TEMPLATE.zephyr',
|
16 |
+
type='xtuner.dataset.map_fns.template_map_fn_factory'),
|
17 |
+
tokenizer=dict(
|
18 |
+
padding_side='right',
|
19 |
+
pretrained_model_name_or_path=
|
20 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
21 |
+
trust_remote_code=True,
|
22 |
+
type='transformers.AutoTokenizer.from_pretrained'),
|
23 |
+
type='xtuner.dataset.process_hf_dataset',
|
24 |
+
use_varlen_attn=False)
|
25 |
+
alpaca_en_path = '/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/datasets--recogna-nlp--ultra-alpaca-ptbr/snapshots/e69900d074177d370a911096fc30bdf407eff666/train.json'
|
26 |
+
batch_size = 2
|
27 |
+
betas = (
|
28 |
+
0.9,
|
29 |
+
0.999,
|
30 |
+
)
|
31 |
+
custom_hooks = [
|
32 |
+
dict(
|
33 |
+
tokenizer=dict(
|
34 |
+
padding_side='right',
|
35 |
+
pretrained_model_name_or_path=
|
36 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
37 |
+
trust_remote_code=True,
|
38 |
+
type='transformers.AutoTokenizer.from_pretrained'),
|
39 |
+
type='xtuner.engine.hooks.DatasetInfoHook'),
|
40 |
+
dict(
|
41 |
+
evaluation_inputs=[
|
42 |
+
'O que é um bode?',
|
43 |
+
'Qual a capital da França?',
|
44 |
+
'Você pode me explicar o Teorema de Pitágoras com um exemplo, por favor?',
|
45 |
+
'Olá, tudo bem? Estou procurando livros de ficção científica para ler, você teria sugestões para mim?',
|
46 |
+
'Resolva a equação de segundo grau x² - x - 30 = 0',
|
47 |
+
'Escreva um código em python para calcular x^y usando divisão e conquista.',
|
48 |
+
],
|
49 |
+
every_n_iters=500,
|
50 |
+
prompt_template='xtuner.utils.PROMPT_TEMPLATE.zephyr',
|
51 |
+
system=
|
52 |
+
'Você é assistente de IA chamado ZephyrBode.\n - O ZephyrBode é um modelo de língua conversacional projetado para ser prestativo, honesto e inofensivo.\n - O ZephyrBode pode entender e se comunicar fluentemente na linguagem escolhida pelo usuário, em especial o Português e o Inglês',
|
53 |
+
tokenizer=dict(
|
54 |
+
padding_side='right',
|
55 |
+
pretrained_model_name_or_path=
|
56 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
57 |
+
trust_remote_code=True,
|
58 |
+
type='transformers.AutoTokenizer.from_pretrained'),
|
59 |
+
type='xtuner.engine.hooks.EvaluateChatHook'),
|
60 |
+
]
|
61 |
+
dataloader_num_workers = 0
|
62 |
+
default_hooks = dict(
|
63 |
+
checkpoint=dict(
|
64 |
+
by_epoch=False,
|
65 |
+
interval=500,
|
66 |
+
max_keep_ckpts=2,
|
67 |
+
type='mmengine.hooks.CheckpointHook'),
|
68 |
+
logger=dict(
|
69 |
+
interval=10,
|
70 |
+
log_metric_by_epoch=False,
|
71 |
+
type='mmengine.hooks.LoggerHook'),
|
72 |
+
param_scheduler=dict(type='mmengine.hooks.ParamSchedulerHook'),
|
73 |
+
sampler_seed=dict(type='mmengine.hooks.DistSamplerSeedHook'),
|
74 |
+
timer=dict(type='mmengine.hooks.IterTimerHook'))
|
75 |
+
env_cfg = dict(
|
76 |
+
cudnn_benchmark=False,
|
77 |
+
dist_cfg=dict(backend='nccl'),
|
78 |
+
mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0))
|
79 |
+
evaluation_freq = 500
|
80 |
+
evaluation_inputs = [
|
81 |
+
'O que é um bode?',
|
82 |
+
'Qual a capital da França?',
|
83 |
+
'Você pode me explicar o Teorema de Pitágoras com um exemplo, por favor?',
|
84 |
+
'Olá, tudo bem? Estou procurando livros de ficção científica para ler, você teria sugestões para mim?',
|
85 |
+
'Resolva a equação de segundo grau x² - x - 30 = 0',
|
86 |
+
'Escreva um código em python para calcular x^y usando divisão e conquista.',
|
87 |
+
]
|
88 |
+
launcher = 'pytorch'
|
89 |
+
load_from = None
|
90 |
+
log_level = 'INFO'
|
91 |
+
log_processor = dict(by_epoch=False)
|
92 |
+
lr = 0.0002
|
93 |
+
max_epochs = 1
|
94 |
+
max_length = 2048
|
95 |
+
max_norm = 1
|
96 |
+
model = dict(
|
97 |
+
llm=dict(
|
98 |
+
pretrained_model_name_or_path=
|
99 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
100 |
+
quantization_config=dict(
|
101 |
+
bnb_4bit_compute_dtype='torch.float16',
|
102 |
+
bnb_4bit_quant_type='nf4',
|
103 |
+
bnb_4bit_use_double_quant=True,
|
104 |
+
llm_int8_has_fp16_weight=False,
|
105 |
+
llm_int8_threshold=6.0,
|
106 |
+
load_in_4bit=True,
|
107 |
+
load_in_8bit=False,
|
108 |
+
type='transformers.BitsAndBytesConfig'),
|
109 |
+
torch_dtype='torch.float16',
|
110 |
+
trust_remote_code=True,
|
111 |
+
type='transformers.AutoModelForCausalLM.from_pretrained'),
|
112 |
+
lora=dict(
|
113 |
+
bias='none',
|
114 |
+
lora_alpha=16,
|
115 |
+
lora_dropout=0.1,
|
116 |
+
r=64,
|
117 |
+
task_type='CAUSAL_LM',
|
118 |
+
type='peft.LoraConfig'),
|
119 |
+
type='xtuner.model.SupervisedFinetune',
|
120 |
+
use_varlen_attn=False)
|
121 |
+
optim_type = 'torch.optim.AdamW'
|
122 |
+
optim_wrapper = dict(
|
123 |
+
optimizer=dict(
|
124 |
+
betas=(
|
125 |
+
0.9,
|
126 |
+
0.999,
|
127 |
+
),
|
128 |
+
lr=0.0002,
|
129 |
+
type='torch.optim.AdamW',
|
130 |
+
weight_decay=0),
|
131 |
+
type='DeepSpeedOptimWrapper')
|
132 |
+
pack_to_max_length = True
|
133 |
+
param_scheduler = [
|
134 |
+
dict(
|
135 |
+
begin=0,
|
136 |
+
by_epoch=True,
|
137 |
+
convert_to_iter_based=True,
|
138 |
+
end=0.03,
|
139 |
+
start_factor=1e-05,
|
140 |
+
type='mmengine.optim.LinearLR'),
|
141 |
+
dict(
|
142 |
+
begin=0.03,
|
143 |
+
by_epoch=True,
|
144 |
+
convert_to_iter_based=True,
|
145 |
+
end=1,
|
146 |
+
eta_min=0.0,
|
147 |
+
type='mmengine.optim.CosineAnnealingLR'),
|
148 |
+
]
|
149 |
+
pretrained_model_name_or_path = '/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473'
|
150 |
+
prompt_template = 'xtuner.utils.PROMPT_TEMPLATE.zephyr'
|
151 |
+
randomness = dict(deterministic=False, seed=None)
|
152 |
+
resume = False
|
153 |
+
runner_type = 'FlexibleRunner'
|
154 |
+
save_steps = 500
|
155 |
+
save_total_limit = 2
|
156 |
+
strategy = dict(
|
157 |
+
config=dict(
|
158 |
+
bf16=dict(enabled=False),
|
159 |
+
fp16=dict(enabled=True, initial_scale_power=16),
|
160 |
+
gradient_accumulation_steps='auto',
|
161 |
+
gradient_clipping='auto',
|
162 |
+
train_micro_batch_size_per_gpu='auto',
|
163 |
+
zero_allow_untested_optimizer=True,
|
164 |
+
zero_force_ds_cpu_optimizer=False,
|
165 |
+
zero_optimization=dict(overlap_comm=True, stage=2)),
|
166 |
+
exclude_frozen_parameters=True,
|
167 |
+
gradient_accumulation_steps=16,
|
168 |
+
gradient_clipping=1,
|
169 |
+
sequence_parallel_size=1,
|
170 |
+
train_micro_batch_size_per_gpu=2,
|
171 |
+
type='xtuner.engine.DeepSpeedStrategy')
|
172 |
+
tokenizer = dict(
|
173 |
+
padding_side='right',
|
174 |
+
pretrained_model_name_or_path=
|
175 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
176 |
+
trust_remote_code=True,
|
177 |
+
type='transformers.AutoTokenizer.from_pretrained')
|
178 |
+
train_cfg = dict(max_epochs=1, type='xtuner.engine.runner.TrainLoop')
|
179 |
+
train_dataloader = dict(
|
180 |
+
batch_size=2,
|
181 |
+
collate_fn=dict(
|
182 |
+
type='xtuner.dataset.collate_fns.default_collate_fn',
|
183 |
+
use_varlen_attn=False),
|
184 |
+
dataset=dict(
|
185 |
+
dataset=dict(
|
186 |
+
data_files=
|
187 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/datasets--recogna-nlp--ultra-alpaca-ptbr/snapshots/e69900d074177d370a911096fc30bdf407eff666/train.json',
|
188 |
+
path='json',
|
189 |
+
type='datasets.load_dataset'),
|
190 |
+
dataset_map_fn='xtuner.dataset.map_fns.ultracabrita_map_fn',
|
191 |
+
max_length=2048,
|
192 |
+
pack_to_max_length=True,
|
193 |
+
remove_unused_columns=True,
|
194 |
+
shuffle_before_pack=True,
|
195 |
+
template_map_fn=dict(
|
196 |
+
template='xtuner.utils.PROMPT_TEMPLATE.zephyr',
|
197 |
+
type='xtuner.dataset.map_fns.template_map_fn_factory'),
|
198 |
+
tokenizer=dict(
|
199 |
+
padding_side='right',
|
200 |
+
pretrained_model_name_or_path=
|
201 |
+
'/petrobr/parceirosbr/home/rafael.rodrigues/.cache/huggingface/hub/models--HuggingFaceH4--zephyr-7b-beta/snapshots/b70e0c9a2d9e14bd1e812d3c398e5f313e93b473',
|
202 |
+
trust_remote_code=True,
|
203 |
+
type='transformers.AutoTokenizer.from_pretrained'),
|
204 |
+
type='xtuner.dataset.process_hf_dataset',
|
205 |
+
use_varlen_attn=False),
|
206 |
+
num_workers=0,
|
207 |
+
sampler=dict(shuffle=True, type='mmengine.dataset.DefaultSampler'))
|
208 |
+
use_varlen_attn = False
|
209 |
+
visualizer = None
|
210 |
+
warmup_ratio = 0.03
|
211 |
+
weight_decay = 0
|
212 |
+
work_dir = './work_dirs/zephyr_7b_beta_qlora_ultraalpaca'
|