Delta-Vector commited on
Commit
15c670e
·
verified ·
1 Parent(s): 45c864b

Upload big.yml

Browse files
Files changed (1) hide show
  1. big.yml +103 -0
big.yml ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ base_model: meta-llama/Llama-3.3-70B-Instruct
2
+ base_model_ignore_patterns: "*/*"
3
+ model_type: AutoModelForCausalLM
4
+ tokenizer_type: AutoTokenizer
5
+
6
+
7
+ hub_model_id: NewEden/BigPicaro
8
+ hub_strategy: "all_checkpoints"
9
+ push_dataset_to_hub:
10
+ hf_use_auth_token: true
11
+
12
+ load_in_8bit: false
13
+ load_in_4bit: false
14
+ strict: false
15
+
16
+
17
+
18
+
19
+ datasets:
20
+ - path: Edens-Gate/Picaro
21
+ type: chat_template
22
+ roles_to_train: ["gpt"]
23
+ field_messages: conversations
24
+ message_field_role: from
25
+ message_field_content: value
26
+ train_on_eos: turn
27
+
28
+ shuffle_merged_datasets: true
29
+ dataset_prepared_path: prepared_data
30
+ val_set_size: 0.0
31
+ output_dir: ./output
32
+
33
+ plugins:
34
+ - axolotl.integrations.liger.LigerPlugin
35
+ - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
36
+ liger_rope: true
37
+ liger_rms_norm: true
38
+ liger_layer_norm: true
39
+ liger_glu_activation: true
40
+ liger_fused_linear_cross_entropy: false
41
+ cut_cross_entropy: true
42
+
43
+ #torch_compile: auto
44
+ #torch_compile_backend:
45
+
46
+ sequence_len: 32768
47
+ sample_packing: true
48
+ eval_sample_packing: false
49
+ pad_to_sequence_len: true
50
+
51
+ adapter: lora
52
+ lora_model_dir:
53
+ lora_r: 128
54
+ lora_alpha: 16
55
+ lora_dropout: 0.05
56
+ lora_target_linear: true
57
+ lora_fan_in_fan_out:
58
+ peft_use_rslora: true
59
+ lora_modules_to_save:
60
+ - embed_tokens
61
+ - lm_head
62
+
63
+ wandb_project: 70b-magnum-lora
64
+ wandb_entity:
65
+ wandb_watch:
66
+ wandb_name:
67
+ wandb_log_model:
68
+
69
+ gradient_accumulation_steps: 1
70
+ micro_batch_size: 2
71
+ num_epochs: 2
72
+ optimizer: paged_ademamix_8bit
73
+ lr_scheduler: cosine
74
+ learning_rate: 4.0e-5
75
+ max_grad_norm: 0.2
76
+
77
+ train_on_inputs: false
78
+ group_by_length: false
79
+ bf16: auto
80
+ fp16:
81
+ tf32: false
82
+
83
+ gradient_checkpointing: unsloth
84
+ early_stopping_patience:
85
+ resume_from_checkpoint:
86
+ local_rank:
87
+ logging_steps: 1
88
+ xformers_attention:
89
+ flash_attention: true
90
+ s2_attention:
91
+
92
+ warmup_steps: 40
93
+ evals_per_epoch:
94
+ eval_table_size:
95
+ eval_max_new_tokens:
96
+ saves_per_epoch: 2
97
+ debug:
98
+ deepspeed: ./deepspeed_configs/zero3_bf16_compile.json
99
+ weight_decay: 0.01
100
+ fsdp:
101
+ fsdp_config:
102
+ special_tokens:
103
+ pad_token: <|finetune_right_pad_id|>