Delta-Vector commited on
Commit
4ae740a
·
verified ·
1 Parent(s): 8d67baf

Upload 3bNewFlavor,yml.yml

Browse files
Files changed (1) hide show
  1. 3bNewFlavor,yml.yml +93 -0
3bNewFlavor,yml.yml ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ base_model: meta-llama/Llama-3.2-3B
2
+ model_type: AutoModelForCausalLM
3
+ tokenizer_type: AutoTokenizer
4
+
5
+ plugins:
6
+ - axolotl.integrations.liger.LigerPlugin
7
+ liger_rope: true
8
+ liger_rms_norm: true
9
+ liger_swiglu: true
10
+ liger_fused_linear_cross_entropy: true
11
+
12
+ load_in_8bit: false
13
+ load_in_3bit: false
14
+ strict: false
15
+
16
+ datasets:
17
+ - path: NewEden/Gryphe-Sonnet3.5-Charcard-Roleplay-unfiltered
18
+ type: sharegpt
19
+ conversation: llama3
20
+ - path: Gryphe/ChatGPT-4o-Writing-Prompts
21
+ type: sharegpt
22
+ conversation: llama3
23
+ - path: Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned
24
+ type: sharegpt
25
+ conversation: llama3
26
+ - path: Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
27
+ type: sharegpt
28
+ conversation: llama3
29
+ - path: NewEden/Gryphe-Sonnet-3.5-35k-Subset
30
+ type: sharegpt
31
+ conversation: llama3
32
+ - path: anthracite-org/kalo_opus_misc_240827
33
+ type: sharegpt
34
+ conversation: llama3
35
+ - path: anthracite-org/kalo_misc_part2
36
+ type: sharegpt
37
+ conversation: llama3
38
+ - path: NewEden/Claude-Instruct-5K
39
+ type: sharegpt
40
+ conversation: llama3
41
+ - path: anthracite-org/kalo-opus-instruct-22k-no-refusal
42
+ type: sharegpt
43
+ conversation: llama3
44
+ chat_template: llama3
45
+ shuffle_merged_datasets: true
46
+ #default_system_message: "You are an assistant that responds to the user."
47
+ dataset_prepared_path: ./holland-3b-data
48
+ val_set_size: 0.0
49
+ output_dir: ./3b-fft-out
50
+
51
+ sequence_len: 32768
52
+ sample_packing: true
53
+ pad_to_sequence_len: true
54
+
55
+ wandb_project: Holland-Mag-3B-LLama
56
+ wandb_entity:
57
+ wandb_watch:
58
+ wandb_name: attempt-llama3-01
59
+ wandb_log_model:
60
+
61
+ gradient_accumulation_steps: 8
62
+ micro_batch_size: 4
63
+ num_epochs: 2
64
+ optimizer: adamw_bnb_8bit
65
+ lr_scheduler: cosine
66
+ learning_rate: 0.000005
67
+
68
+ train_on_inputs: false
69
+ group_by_length: false
70
+ bf16: auto
71
+ fp16:
72
+ tf32: false
73
+
74
+ gradient_checkpointing: true
75
+ early_stopping_patience:
76
+ resume_from_checkpoint:
77
+ local_rank:
78
+ logging_steps: 1
79
+ xformers_attention:
80
+ flash_attention: true
81
+
82
+ warmup_steps: 10
83
+ evals_per_epoch:
84
+ eval_table_size:
85
+ eval_max_new_tokens:
86
+ saves_per_epoch: 2
87
+ debug:
88
+ deepspeed:
89
+ weight_decay: 0.01
90
+ fsdp:
91
+ fsdp_config:
92
+ special_tokens:
93
+ pad_token: <|finetune_right_pad_id|>