Delta-Vector commited on
Commit
41672ff
·
verified ·
1 Parent(s): c01b999

Upload 2 files

Browse files
Files changed (2) hide show
  1. 4B-Style.yml +84 -0
  2. magnum.yml +81 -0
4B-Style.yml ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ base_model: NewEden/Hamanasu-KTO-V2
2
+ model_type: AutoModelForCausalLM
3
+ tokenizer_type: AutoTokenizer
4
+
5
+ load_in_8bit: false
6
+ load_in_4bit: false
7
+ strict: false
8
+
9
+ datasets:
10
+ - path: PocketDoc/Dans-Personamaxx-VN
11
+ type: dan-chat-advanced
12
+ - path: PocketDoc/Dans-Prosemaxx-Gutenberg
13
+ type: dan-chat-advanced
14
+ - path: NewEden/Light-Novels-ShareGPT
15
+ type: dan-chat-advanced
16
+ - path: NewEden/OpenCAI-ShareGPT
17
+ type: dan-chat-advanced
18
+ - path: NewEden/Wata-Oshi-LNs-ShareGPT-Final
19
+ type: dan-chat-advanced
20
+ - path: NewEden/DeepseekRP-Filtered
21
+ type: dan-chat-advanced
22
+ - path: NewEden/Storium-Prefixed-Clean
23
+ type: dan-chat-advanced
24
+ - path: NewEden/PIPPA-Mega-Filtered
25
+ type: dan-chat-advanced
26
+ - path: NewEden/Orion-Misc-Sharegpt-Prefixed
27
+ type: dan-chat-advanced
28
+ - path: NewEden/LIMARP-Complexity
29
+ type: dan-chat-advanced
30
+ val_set_size: 0.01
31
+ output_dir: ./outputs/out
32
+
33
+ adapter:
34
+ lora_r:
35
+ lora_alpha:
36
+ lora_dropout:
37
+ lora_target_linear:
38
+
39
+ sequence_len: 32768
40
+ sample_packing: true
41
+ eval_sample_packing: false
42
+ pad_to_sequence_len: true
43
+
44
+ wandb_project: tavbussy
45
+ wandb_entity:
46
+ wandb_watch:
47
+ wandb_name: RP-Ussy-attempt-01
48
+ wandb_log_model:
49
+
50
+ gradient_accumulation_steps: 4
51
+ micro_batch_size: 4
52
+ num_epochs: 4
53
+ optimizer: adamw_bnb_8bit
54
+ lr_scheduler: cosine
55
+ learning_rate: 0.00001
56
+ weight_decay: 0.02
57
+
58
+ train_on_inputs: false
59
+ group_by_length: false
60
+ bf16: auto
61
+ fp16:
62
+ tf32: true
63
+
64
+ gradient_checkpointing: true
65
+ early_stopping_patience:
66
+ resume_from_checkpoint:
67
+ local_rank:
68
+ logging_steps: 1
69
+ xformers_attention:
70
+ flash_attention: true
71
+
72
+ warmup_ratio: 0.1
73
+ evals_per_epoch: 4
74
+ eval_table_size:
75
+ eval_max_new_tokens: 128
76
+ saves_per_epoch: 1
77
+
78
+ debug:
79
+ deepspeed:
80
+ fsdp:
81
+ fsdp_config:
82
+
83
+ special_tokens:
84
+ pad_token: <|finetune_right_pad_id|>
magnum.yml ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ base_model: NewEden/Hamanasu-KTO-V2
2
+ model_type: AutoModelForCausalLM
3
+ tokenizer_type: AutoTokenizer
4
+
5
+ load_in_8bit: false
6
+ load_in_4bit: false
7
+ strict: false
8
+
9
+ datasets:
10
+ - path: PocketDoc/Dans-Personamaxx-Logs
11
+ type: dan-chat-advanced
12
+ - path: anthracite-org/kalo-opus-instruct-22k-no-refusal
13
+ type: dan-chat-advanced
14
+ - path: lodrick-the-lafted/kalo-opus-instruct-3k-filtered
15
+ type: dan-chat-advanced
16
+ - path: anthracite-org/nopm_claude_writing_fixed
17
+ type: dan-chat-advanced
18
+ - path: anthracite-org/kalo_opus_misc_240827
19
+ type: dan-chat-advanced
20
+ - path: anthracite-org/kalo_misc_part2
21
+ type: dan-chat-advanced
22
+ - path: NewEden/Claude-Instruct-5K
23
+ type: dan-chat-advanced
24
+ - path: NewEden/Claude-Instruct-2.7K
25
+ type: dan-chat-advanced
26
+
27
+ val_set_size: 0.01
28
+ output_dir: ./outputs/out
29
+
30
+ adapter:
31
+ lora_r:
32
+ lora_alpha:
33
+ lora_dropout:
34
+ lora_target_linear:
35
+
36
+ sequence_len: 32768
37
+ sample_packing: true
38
+ eval_sample_packing: false
39
+ pad_to_sequence_len: true
40
+
41
+ wandb_project: tavbussy
42
+ wandb_entity:
43
+ wandb_watch:
44
+ wandb_name: magnum-attempt-01
45
+ wandb_log_model:
46
+
47
+ gradient_accumulation_steps: 4
48
+ micro_batch_size: 4
49
+ num_epochs: 2
50
+ optimizer: adamw_bnb_8bit
51
+ lr_scheduler: cosine
52
+ learning_rate: 0.00002
53
+ weight_decay: 0.05
54
+
55
+ train_on_inputs: false
56
+ group_by_length: false
57
+ bf16: auto
58
+ fp16:
59
+ tf32: true
60
+
61
+ gradient_checkpointing: true
62
+ early_stopping_patience:
63
+ resume_from_checkpoint:
64
+ local_rank:
65
+ logging_steps: 1
66
+ xformers_attention:
67
+ flash_attention: true
68
+
69
+ warmup_ratio: 0.1
70
+ evals_per_epoch: 4
71
+ eval_table_size:
72
+ eval_max_new_tokens: 128
73
+ saves_per_epoch: 1
74
+
75
+ debug:
76
+ deepspeed:
77
+ fsdp:
78
+ fsdp_config:
79
+
80
+ special_tokens:
81
+ pad_token: <|finetune_right_pad_id|>