jfarland commited on
Commit
4c20c67
1 Parent(s): 1a5f356

Upload cfg.yaml

Browse files
Files changed (1) hide show
  1. cfg.yaml +102 -0
cfg.yaml ADDED
@@ -0,0 +1,102 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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: ''
8
+ augmentation:
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: true
14
+ add_eos_token_to_prompt: true
15
+ add_eos_token_to_system: true
16
+ answer_column: content
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
+ - title
29
+ system_column: None
30
+ text_answer_separator: <|answer|>
31
+ text_prompt_start: <|prompt|>
32
+ text_system_start: <|system|>
33
+ train_dataframe: /home/fsuser/projects/h2o-llmstudio/data/user/influencers_data_prepared/influencers_data_prepared.csv
34
+ validation_dataframe: None
35
+ validation_size: 0.01
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
+ huggingface_branch: main
46
+ mixed_precision: true
47
+ number_of_workers: 8
48
+ seed: -1
49
+ trust_remote_code: true
50
+ use_deepspeed: false
51
+ experiment_name: foamy-oriole
52
+ llm_backbone: h2oai/h2ogpt-4096-llama2-7b
53
+ logging:
54
+ logger: None
55
+ neptune_project: ''
56
+ output_directory: /home/fsuser/projects/h2o-llmstudio/output/user/foamy-oriole/
57
+ prediction:
58
+ batch_size_inference: 0
59
+ do_sample: false
60
+ max_length_inference: 256
61
+ metric: BLEU
62
+ metric_gpt_model: gpt-3.5-turbo-0301
63
+ min_length_inference: 2
64
+ num_beams: 1
65
+ num_history: 4
66
+ repetition_penalty: 1.2
67
+ stop_tokens: ''
68
+ temperature: 0.0
69
+ top_k: 0
70
+ top_p: 1.0
71
+ problem_type: text_causal_language_modeling
72
+ tokenizer:
73
+ add_prefix_space: false
74
+ add_prompt_answer_tokens: false
75
+ max_length: 512
76
+ max_length_answer: 256
77
+ max_length_prompt: 256
78
+ padding_quantile: 1.0
79
+ use_fast: true
80
+ training:
81
+ batch_size: 2
82
+ differential_learning_rate: 1.0e-05
83
+ differential_learning_rate_layers: []
84
+ drop_last_batch: true
85
+ epochs: 1
86
+ evaluate_before_training: false
87
+ evaluation_epochs: 1.0
88
+ grad_accumulation: 1
89
+ gradient_clip: 0.0
90
+ learning_rate: 0.0001
91
+ lora: true
92
+ lora_alpha: 16
93
+ lora_dropout: 0.05
94
+ lora_r: 4
95
+ lora_target_modules: ''
96
+ loss_function: TokenAveragedCrossEntropy
97
+ optimizer: AdamW
98
+ save_best_checkpoint: false
99
+ schedule: Cosine
100
+ train_validation_data: false
101
+ warmup_epochs: 0.0
102
+ weight_decay: 0.0