zakariarada commited on
Commit
dec739d
·
verified ·
1 Parent(s): 48474ab

Upload cfg.yaml

Browse files
Files changed (1) hide show
  1. cfg.yaml +107 -0
cfg.yaml ADDED
@@ -0,0 +1,107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ architecture:
2
+ backbone_dtype: int4
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: 0.0
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:
18
+ - label
19
+ chatbot_author: H2O.ai
20
+ chatbot_name: h2oGPT
21
+ data_sample: 1.0
22
+ data_sample_choice:
23
+ - Train
24
+ - Validation
25
+ limit_chained_samples: false
26
+ mask_prompt_labels: true
27
+ num_classes: 1
28
+ only_last_answer: false
29
+ parent_id_column: None
30
+ personalize: false
31
+ prompt_column:
32
+ - text
33
+ prompt_column_separator: \n\n
34
+ system_column: None
35
+ text_answer_separator: ''
36
+ text_prompt_start: ''
37
+ text_system_start: ''
38
+ train_dataframe: /home/hunterx/h2o-llmstudio/data/user/imdb/train.pq
39
+ validation_dataframe: None
40
+ validation_size: 0.01
41
+ validation_strategy: automatic
42
+ environment:
43
+ compile_model: false
44
+ deepspeed_allgather_bucket_size: 1000000
45
+ deepspeed_method: ZeRO2
46
+ deepspeed_reduce_bucket_size: 1000000
47
+ deepspeed_stage3_param_persistence_threshold: 1000000
48
+ deepspeed_stage3_prefetch_bucket_size: 1000000
49
+ find_unused_parameters: false
50
+ gpus:
51
+ - '0'
52
+ huggingface_branch: main
53
+ mixed_precision: true
54
+ mixed_precision_dtype: bfloat16
55
+ number_of_workers: 8
56
+ seed: -1
57
+ trust_remote_code: true
58
+ use_deepspeed: false
59
+ experiment_name: onyx-mastiff
60
+ llm_backbone: h2oai/h2o-danube3-500m-chat
61
+ logging:
62
+ log_all_ranks: false
63
+ log_step_size: absolute
64
+ logger: None
65
+ neptune_project: ''
66
+ wandb_entity: ''
67
+ wandb_project: ''
68
+ output_directory: /home/hunterx/h2o-llmstudio/output/user/onyx-mastiff/
69
+ prediction:
70
+ batch_size_inference: 0
71
+ metric: AUC
72
+ problem_type: text_causal_classification_modeling
73
+ tokenizer:
74
+ add_prompt_answer_tokens: false
75
+ max_length: 512
76
+ padding_quantile: 1.0
77
+ tokenizer_kwargs: '{"use_fast": true, "add_prefix_space": false}'
78
+ training:
79
+ attention_implementation: auto
80
+ batch_size: 2
81
+ differential_learning_rate: 1.0e-05
82
+ differential_learning_rate_layers:
83
+ - classification_head
84
+ drop_last_batch: true
85
+ epochs: 1
86
+ evaluate_before_training: false
87
+ evaluation_epochs: 1.0
88
+ freeze_layers: []
89
+ grad_accumulation: 1
90
+ gradient_clip: 0.0
91
+ learning_rate: 0.0001
92
+ lora: true
93
+ lora_alpha: 16
94
+ lora_dropout: 0.05
95
+ lora_r: 4
96
+ lora_target_modules: ''
97
+ lora_unfreeze_layers: []
98
+ loss_function: BinaryCrossEntropyLoss
99
+ min_learning_rate_ratio: 0.0
100
+ optimizer: AdamW
101
+ save_checkpoint: last
102
+ schedule: Cosine
103
+ train_validation_data: false
104
+ use_dora: false
105
+ use_rslora: false
106
+ warmup_epochs: 0.0
107
+ weight_decay: 0.0