samvelkoch
commited on
Upload cfg.yaml
Browse files
cfg.yaml
ADDED
@@ -0,0 +1,227 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
architecture:
|
2 |
+
backbone_dtype: float32
|
3 |
+
gradient_checkpointing: true
|
4 |
+
intermediate_dropout: 0.0
|
5 |
+
pretrained: true
|
6 |
+
pretrained_weights: ''
|
7 |
+
augmentation:
|
8 |
+
neftune_noise_alpha: 0.0
|
9 |
+
random_parent_probability: 0.0
|
10 |
+
skip_parent_probability: 0.0
|
11 |
+
token_mask_probability: 0.1
|
12 |
+
dataset:
|
13 |
+
add_eos_token_to_answer: false
|
14 |
+
add_eos_token_to_prompt: false
|
15 |
+
add_eos_token_to_system: false
|
16 |
+
add_prompt_answer_tokens: false
|
17 |
+
answer_column: score
|
18 |
+
chatbot_author: H2O.ai
|
19 |
+
chatbot_name: h2oGPT
|
20 |
+
data_sample: 1.0
|
21 |
+
data_sample_choice:
|
22 |
+
- Train
|
23 |
+
- Validation
|
24 |
+
limit_chained_samples: false
|
25 |
+
mask_prompt_labels: true
|
26 |
+
num_classes: 6
|
27 |
+
parent_id_column: None
|
28 |
+
personalize: false
|
29 |
+
prompt_column:
|
30 |
+
- full_text
|
31 |
+
- count_sentence
|
32 |
+
- count_word
|
33 |
+
- count_paragraph
|
34 |
+
- count_symbol
|
35 |
+
- count_punctuation
|
36 |
+
- count_stop_words
|
37 |
+
- count_ner
|
38 |
+
- CARDINAL
|
39 |
+
- DATE
|
40 |
+
- EVENT
|
41 |
+
- FAC
|
42 |
+
- GPE
|
43 |
+
- LANGUAGE
|
44 |
+
- LAW
|
45 |
+
- LOC
|
46 |
+
- MONEY
|
47 |
+
- NORP
|
48 |
+
- ORDINAL
|
49 |
+
- ORG
|
50 |
+
- PERCENT
|
51 |
+
- PERSON
|
52 |
+
- PRODUCT
|
53 |
+
- QUANTITY
|
54 |
+
- TIME
|
55 |
+
- WORK_OF_ART
|
56 |
+
- mean_ner_sentence
|
57 |
+
- len_ner
|
58 |
+
- mean_ner
|
59 |
+
- proc_ner_per_text
|
60 |
+
- max_ner_per_sentence
|
61 |
+
- independent_clauses
|
62 |
+
- dependent_clauses
|
63 |
+
- formality_level
|
64 |
+
- simple_sentence_count
|
65 |
+
- complex_sentence_count
|
66 |
+
- has_intro
|
67 |
+
- count_intro
|
68 |
+
- has_repetitions
|
69 |
+
- count_repetitions
|
70 |
+
- coherence
|
71 |
+
- -LRB-
|
72 |
+
- -RRB-
|
73 |
+
- ADD
|
74 |
+
- AFX
|
75 |
+
- CC
|
76 |
+
- CD
|
77 |
+
- DT
|
78 |
+
- EX
|
79 |
+
- FW
|
80 |
+
- HYPH
|
81 |
+
- IN
|
82 |
+
- JJ
|
83 |
+
- JJR
|
84 |
+
- JJS
|
85 |
+
- LS
|
86 |
+
- MD
|
87 |
+
- NFP
|
88 |
+
- NN
|
89 |
+
- NNP
|
90 |
+
- NNPS
|
91 |
+
- NNS
|
92 |
+
- PDT
|
93 |
+
- POS
|
94 |
+
- PRP
|
95 |
+
- PRP$
|
96 |
+
- RB
|
97 |
+
- RBR
|
98 |
+
- RBS
|
99 |
+
- RP
|
100 |
+
- SYM
|
101 |
+
- TO
|
102 |
+
- UH
|
103 |
+
- VB
|
104 |
+
- VBD
|
105 |
+
- VBG
|
106 |
+
- VBN
|
107 |
+
- VBP
|
108 |
+
- VBZ
|
109 |
+
- WDT
|
110 |
+
- WP
|
111 |
+
- WP$
|
112 |
+
- WRB
|
113 |
+
- XX
|
114 |
+
- _SP
|
115 |
+
- ROOT
|
116 |
+
- acl
|
117 |
+
- acomp
|
118 |
+
- advcl
|
119 |
+
- advmod
|
120 |
+
- agent
|
121 |
+
- amod
|
122 |
+
- appos
|
123 |
+
- attr
|
124 |
+
- aux
|
125 |
+
- auxpass
|
126 |
+
- case
|
127 |
+
- cc
|
128 |
+
- ccomp
|
129 |
+
- compound
|
130 |
+
- conj
|
131 |
+
- csubj
|
132 |
+
- csubjpass
|
133 |
+
- dative
|
134 |
+
- dep
|
135 |
+
- det
|
136 |
+
- dobj
|
137 |
+
- expl
|
138 |
+
- intj
|
139 |
+
- mark
|
140 |
+
- meta
|
141 |
+
- neg
|
142 |
+
- nmod
|
143 |
+
- npadvmod
|
144 |
+
- nsubj
|
145 |
+
- nsubjpass
|
146 |
+
- nummod
|
147 |
+
- oprd
|
148 |
+
- parataxis
|
149 |
+
- pcomp
|
150 |
+
- pobj
|
151 |
+
- poss
|
152 |
+
- preconj
|
153 |
+
- predet
|
154 |
+
- prep
|
155 |
+
- prt
|
156 |
+
- punct
|
157 |
+
- quantmod
|
158 |
+
- relcl
|
159 |
+
- xcomp
|
160 |
+
system_column: None
|
161 |
+
text_answer_separator: ''
|
162 |
+
text_prompt_start: ''
|
163 |
+
text_system_start: ''
|
164 |
+
train_dataframe: /root/h2o-llmstudio/data/user/train_spacy_mamba/train_spacy_mamba.csv
|
165 |
+
validation_dataframe: None
|
166 |
+
validation_size: 0.01
|
167 |
+
validation_strategy: automatic
|
168 |
+
environment:
|
169 |
+
compile_model: false
|
170 |
+
deepspeed_allgather_bucket_size: 1000000
|
171 |
+
deepspeed_method: ZeRO2
|
172 |
+
deepspeed_reduce_bucket_size: 1000000
|
173 |
+
deepspeed_stage3_param_persistence_threshold: 1000000
|
174 |
+
deepspeed_stage3_prefetch_bucket_size: 1000000
|
175 |
+
find_unused_parameters: false
|
176 |
+
gpus:
|
177 |
+
- '0'
|
178 |
+
huggingface_branch: main
|
179 |
+
mixed_precision: true
|
180 |
+
mixed_precision_dtype: bfloat16
|
181 |
+
number_of_workers: 8
|
182 |
+
seed: -1
|
183 |
+
trust_remote_code: true
|
184 |
+
use_deepspeed: false
|
185 |
+
experiment_name: spacy-mamba
|
186 |
+
llm_backbone: h2oai/h2ogpt-4096-llama2-7b
|
187 |
+
logging:
|
188 |
+
logger: Neptune
|
189 |
+
neptune_project: samvelkoch/essay
|
190 |
+
output_directory: /root/h2o-llmstudio/output/user/spacy-mamba/
|
191 |
+
prediction:
|
192 |
+
batch_size_inference: 0
|
193 |
+
metric: Accuracy
|
194 |
+
problem_type: text_causal_classification_modeling
|
195 |
+
tokenizer:
|
196 |
+
add_prompt_answer_tokens: false
|
197 |
+
max_length: 10240
|
198 |
+
padding_quantile: 1.0
|
199 |
+
tokenizer_kwargs: '{"use_fast": true, "add_prefix_space": false}'
|
200 |
+
training:
|
201 |
+
batch_size: 2
|
202 |
+
differential_learning_rate: 1.0e-05
|
203 |
+
differential_learning_rate_layers:
|
204 |
+
- classification_head
|
205 |
+
drop_last_batch: true
|
206 |
+
epochs: 1
|
207 |
+
evaluate_before_training: false
|
208 |
+
evaluation_epochs: 1.0
|
209 |
+
freeze_layers: []
|
210 |
+
grad_accumulation: 1
|
211 |
+
gradient_clip: 0.0
|
212 |
+
learning_rate: 0.0001
|
213 |
+
lora: true
|
214 |
+
lora_alpha: 16
|
215 |
+
lora_dropout: 0.05
|
216 |
+
lora_r: 4
|
217 |
+
lora_target_modules: ''
|
218 |
+
lora_unfreeze_layers: []
|
219 |
+
loss_function: CrossEntropyLoss
|
220 |
+
optimizer: AdamW
|
221 |
+
save_checkpoint: last
|
222 |
+
schedule: Cosine
|
223 |
+
train_validation_data: false
|
224 |
+
use_dora: false
|
225 |
+
use_flash_attention_2: true
|
226 |
+
warmup_epochs: 0.0
|
227 |
+
weight_decay: 0.0
|