SaffalPoosh commited on
Commit
1d2242f
1 Parent(s): 3c9dd0a

Upload cfg.yaml

Browse files
Files changed (1) hide show
  1. cfg.yaml +105 -0
cfg.yaml ADDED
@@ -0,0 +1,105 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ architecture:
2
+ backbone_dtype: int4
3
+ force_embedding_gradients: false
4
+ gradient_checkpointing: false
5
+ intermediate_dropout: 0.0
6
+ pretrained: true
7
+ pretrained_weights: ''
8
+ augmentation:
9
+ random_parent_probability: 0.05
10
+ skip_parent_probability: 0.05
11
+ token_mask_probability: 0.05
12
+ dataset:
13
+ add_eos_token_to_answer: true
14
+ add_eos_token_to_prompt: true
15
+ add_eos_token_to_system: true
16
+ answer_column: response
17
+ chatbot_author: H2O.ai
18
+ chatbot_name: h2oGPT
19
+ data_sample: 1.0
20
+ data_sample_choice:
21
+ - Train
22
+ - Validation
23
+ limit_chained_samples: false
24
+ mask_prompt_labels: true
25
+ parent_id_column: None
26
+ personalize: false
27
+ prompt_column:
28
+ - prompt
29
+ system_column: None
30
+ text_answer_separator: <|answer|>
31
+ text_prompt_start: <|prompt|>
32
+ text_system_start: <|system|>
33
+ train_dataframe: /workspace/data/user/ultimus_2K/ultimus_2K.csv
34
+ validation_dataframe: None
35
+ validation_size: 0.1
36
+ validation_strategy: automatic
37
+ environment:
38
+ compile_model: false
39
+ deepspeed_reduce_bucket_size: 1000000
40
+ deepspeed_stage3_param_persistence_threshold: 1000000
41
+ deepspeed_stage3_prefetch_bucket_size: 1000000
42
+ find_unused_parameters: false
43
+ gpus:
44
+ - '0'
45
+ - '1'
46
+ - '2'
47
+ - '3'
48
+ huggingface_branch: main
49
+ mixed_precision: true
50
+ number_of_workers: 8
51
+ seed: -1
52
+ trust_remote_code: true
53
+ use_deepspeed: false
54
+ experiment_name: rampant-kagu.1.1.1.1
55
+ llm_backbone: mistralai/Mistral-7B-v0.1
56
+ logging:
57
+ logger: None
58
+ neptune_project: ''
59
+ output_directory: /workspace/output/user/rampant-kagu.1.1.1.1/
60
+ prediction:
61
+ batch_size_inference: 4
62
+ do_sample: true
63
+ max_length_inference: 256
64
+ metric: BLEU
65
+ metric_gpt_model: gpt-4
66
+ min_length_inference: 2
67
+ num_beams: 2
68
+ num_history: 4
69
+ repetition_penalty: 1.2
70
+ stop_tokens: ''
71
+ temperature: 0.35
72
+ top_k: 0
73
+ top_p: 1.0
74
+ problem_type: text_causal_language_modeling
75
+ tokenizer:
76
+ add_prefix_space: false
77
+ add_prompt_answer_tokens: false
78
+ max_length: 704
79
+ max_length_answer: 352
80
+ max_length_prompt: 256
81
+ padding_quantile: 1.0
82
+ use_fast: true
83
+ training:
84
+ batch_size: 4
85
+ differential_learning_rate: 1.0e-05
86
+ differential_learning_rate_layers: []
87
+ drop_last_batch: true
88
+ epochs: 4
89
+ evaluate_before_training: false
90
+ evaluation_epochs: 1.0
91
+ grad_accumulation: 1
92
+ gradient_clip: 1.0
93
+ learning_rate: 1.0e-05
94
+ lora: true
95
+ lora_alpha: 16
96
+ lora_dropout: 0.05
97
+ lora_r: 8
98
+ lora_target_modules: ''
99
+ loss_function: TokenAveragedCrossEntropy
100
+ optimizer: AdamW
101
+ save_best_checkpoint: false
102
+ schedule: Cosine
103
+ train_validation_data: false
104
+ warmup_epochs: 1.0
105
+ weight_decay: 1.0e-05