Upload cfg.yaml
Browse files
cfg.yaml
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
architecture:
|
2 |
+
backbone_dtype: int4
|
3 |
+
force_embedding_gradients: false
|
4 |
+
gradient_checkpointing: true
|
5 |
+
intermediate_dropout: 0.0
|
6 |
+
pretrained: true
|
7 |
+
pretrained_weights: /workspace/h2o-llmstudio/output/user/taupe-bumblebee.1/checkpoint.pth
|
8 |
+
augmentation:
|
9 |
+
neftune_noise_alpha: 0.0
|
10 |
+
random_parent_probability: 0.0
|
11 |
+
skip_parent_probability: 0.0
|
12 |
+
token_mask_probability: 0.0
|
13 |
+
dataset:
|
14 |
+
add_eos_token_to_answer: true
|
15 |
+
add_eos_token_to_prompt: true
|
16 |
+
add_eos_token_to_system: true
|
17 |
+
answer_column: assistant
|
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 |
+
parent_id_column: parent_id
|
27 |
+
personalize: false
|
28 |
+
prompt_column:
|
29 |
+
- user
|
30 |
+
system_column: system
|
31 |
+
text_answer_separator: <|answer|>
|
32 |
+
text_prompt_start: <|prompt|>
|
33 |
+
text_system_start: <|system|>
|
34 |
+
train_dataframe: /workspace/h2o-llmstudio/data/user/dataaug_train_v1/dataaug_train_v1.parquet
|
35 |
+
validation_dataframe: /workspace/h2o-llmstudio/data/user/dataaug_train_v1/dataaug_val_v1.parquet
|
36 |
+
validation_size: 0.01
|
37 |
+
validation_strategy: custom
|
38 |
+
environment:
|
39 |
+
compile_model: false
|
40 |
+
deepspeed_allgather_bucket_size: 1000000
|
41 |
+
deepspeed_method: ZeRO2
|
42 |
+
deepspeed_reduce_bucket_size: 1000000
|
43 |
+
deepspeed_stage3_param_persistence_threshold: 1000000
|
44 |
+
deepspeed_stage3_prefetch_bucket_size: 1000000
|
45 |
+
find_unused_parameters: false
|
46 |
+
gpus:
|
47 |
+
- '0'
|
48 |
+
huggingface_branch: main
|
49 |
+
mixed_precision: true
|
50 |
+
mixed_precision_dtype: bfloat16
|
51 |
+
number_of_workers: 8
|
52 |
+
seed: -1
|
53 |
+
trust_remote_code: true
|
54 |
+
use_deepspeed: false
|
55 |
+
experiment_name: taupe-bumblebee.1.1
|
56 |
+
llm_backbone: google/gemma-2b
|
57 |
+
logging:
|
58 |
+
logger: None
|
59 |
+
neptune_project: ''
|
60 |
+
output_directory: /workspace/h2o-llmstudio/output/user/taupe-bumblebee.1.1/
|
61 |
+
prediction:
|
62 |
+
batch_size_inference: 0
|
63 |
+
do_sample: false
|
64 |
+
max_length_inference: 256
|
65 |
+
max_time: 0.0
|
66 |
+
metric: BLEU
|
67 |
+
metric_gpt_model: gpt-3.5-turbo-0301
|
68 |
+
metric_gpt_template: general
|
69 |
+
min_length_inference: 2
|
70 |
+
num_beams: 1
|
71 |
+
num_history: 4
|
72 |
+
repetition_penalty: 1.0
|
73 |
+
stop_tokens: ''
|
74 |
+
temperature: 0.0
|
75 |
+
top_k: 0
|
76 |
+
top_p: 1.0
|
77 |
+
problem_type: text_causal_language_modeling
|
78 |
+
tokenizer:
|
79 |
+
add_prompt_answer_tokens: false
|
80 |
+
max_length: 2560
|
81 |
+
max_length_answer: 3232
|
82 |
+
max_length_prompt: 2208
|
83 |
+
padding_quantile: 1.0
|
84 |
+
use_fast: true
|
85 |
+
training:
|
86 |
+
batch_size: 2
|
87 |
+
differential_learning_rate: 1.0e-05
|
88 |
+
differential_learning_rate_layers: []
|
89 |
+
drop_last_batch: true
|
90 |
+
epochs: 1
|
91 |
+
evaluate_before_training: false
|
92 |
+
evaluation_epochs: 1.0
|
93 |
+
grad_accumulation: 1
|
94 |
+
gradient_clip: 0.0
|
95 |
+
learning_rate: 0.0001
|
96 |
+
lora: true
|
97 |
+
lora_alpha: 16
|
98 |
+
lora_dropout: 0.05
|
99 |
+
lora_r: 4
|
100 |
+
lora_target_modules: ''
|
101 |
+
loss_function: TokenAveragedCrossEntropy
|
102 |
+
optimizer: AdamW
|
103 |
+
save_checkpoint: last
|
104 |
+
schedule: Cosine
|
105 |
+
train_validation_data: false
|
106 |
+
use_dora: false
|
107 |
+
use_flash_attention_2: false
|
108 |
+
warmup_epochs: 0.0
|
109 |
+
weight_decay: 0.0
|