Anim Mina commited on
Commit
1e1b4b4
1 Parent(s): c4fb3f4

End of training

Browse files
Files changed (2) hide show
  1. README.md +167 -0
  2. adapter_model.bin +3 -0
README.md ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ library_name: peft
4
+ tags:
5
+ - axolotl
6
+ - generated_from_trainer
7
+ base_model: cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser
8
+ model-index:
9
+ - name: cheater-7b
10
+ results: []
11
+ ---
12
+
13
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
14
+ should probably proofread and complete it, then remove this comment. -->
15
+
16
+ [<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
17
+ <details><summary>See axolotl config</summary>
18
+
19
+ axolotl version: `0.4.0`
20
+ ```yaml
21
+ base_model: cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser
22
+ model_type: MistralForCausalLM
23
+ tokenizer_type: LlamaTokenizer
24
+
25
+ load_in_8bit: true
26
+ load_in_4bit: false
27
+ strict: false
28
+
29
+ datasets:
30
+ - path: ./julia/data.jsonl
31
+ type: sharegpt
32
+ conversation: chatml
33
+ dataset_prepared_path: ./julia/prepared_data
34
+ chat_template: chatml
35
+ val_set_size: 0.05
36
+ output_dir: ./julia/lora-out
37
+ hub_model_id: animmina/cheater-7b
38
+ hub_strategy: every_save
39
+ hf_use_auth_token: true
40
+
41
+ sequence_len: 2048
42
+ sample_packing: true
43
+ pad_to_sequence_len: true
44
+ eval_sample_packing: false
45
+
46
+ adapter: lora
47
+ lora_model_dir:
48
+ lora_r: 8
49
+ lora_alpha: 16
50
+ lora_dropout: 0.05
51
+ lora_target_linear: true
52
+ lora_target_modules:
53
+ - gate_proj
54
+ - down_proj
55
+ - up_proj
56
+ - q_proj
57
+ - v_proj
58
+ - k_proj
59
+ - o_proj
60
+
61
+ wandb_project: cheater-7b
62
+ wandb_entity:
63
+ wandb_watch:
64
+ wandb_name: v02
65
+ wandb_log_model:
66
+
67
+ gradient_accumulation_steps: 1
68
+ micro_batch_size: 4
69
+ num_epochs: 4
70
+ optimizer: adamw_bnb_8bit
71
+ lr_scheduler: cosine
72
+ learning_rate: 0.00003
73
+
74
+ train_on_inputs: false
75
+ group_by_length: false
76
+ bf16: auto
77
+ fp16:
78
+ tf32: false
79
+
80
+ gradient_checkpointing: true
81
+ early_stopping_patience:
82
+ resume_from_checkpoint:
83
+ local_rank:
84
+ logging_steps: 1
85
+ xformers_attention:
86
+ flash_attention: true
87
+
88
+ warmup_steps: 10
89
+ evals_per_epoch: 4
90
+ eval_table_size:
91
+ eval_max_new_tokens: 128
92
+ saves_per_epoch: 1
93
+ debug:
94
+ deepspeed:
95
+ weight_decay: 0.0
96
+ fsdp:
97
+ fsdp_config:
98
+ special_tokens:
99
+ bos_token: "<s>"
100
+ eos_token: "<|im_end|>"
101
+ unk_token: "<unk>"
102
+
103
+ ```
104
+
105
+ </details><br>
106
+
107
+ # cheater-7b
108
+
109
+ This model is a fine-tuned version of [cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser](https://huggingface.co/cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser) on the None dataset.
110
+ It achieves the following results on the evaluation set:
111
+ - Loss: 0.5741
112
+
113
+ ## Model description
114
+
115
+ More information needed
116
+
117
+ ## Intended uses & limitations
118
+
119
+ More information needed
120
+
121
+ ## Training and evaluation data
122
+
123
+ More information needed
124
+
125
+ ## Training procedure
126
+
127
+ ### Training hyperparameters
128
+
129
+ The following hyperparameters were used during training:
130
+ - learning_rate: 3e-05
131
+ - train_batch_size: 4
132
+ - eval_batch_size: 4
133
+ - seed: 42
134
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
135
+ - lr_scheduler_type: cosine
136
+ - lr_scheduler_warmup_steps: 10
137
+ - num_epochs: 4
138
+
139
+ ### Training results
140
+
141
+ | Training Loss | Epoch | Step | Validation Loss |
142
+ |:-------------:|:-----:|:----:|:---------------:|
143
+ | 0.554 | 0.04 | 1 | 0.6521 |
144
+ | 0.4152 | 0.26 | 7 | 0.6499 |
145
+ | 0.3984 | 0.52 | 14 | 0.6283 |
146
+ | 0.4133 | 0.78 | 21 | 0.6140 |
147
+ | 0.3772 | 1.04 | 28 | 0.5951 |
148
+ | 0.3855 | 1.22 | 35 | 0.5869 |
149
+ | 0.4077 | 1.48 | 42 | 0.5840 |
150
+ | 0.3104 | 1.74 | 49 | 0.5793 |
151
+ | 0.3345 | 2.0 | 56 | 0.5776 |
152
+ | 0.3207 | 2.19 | 63 | 0.5761 |
153
+ | 0.3679 | 2.44 | 70 | 0.5784 |
154
+ | 0.3593 | 2.7 | 77 | 0.5781 |
155
+ | 0.2391 | 2.96 | 84 | 0.5761 |
156
+ | 0.3329 | 3.15 | 91 | 0.5743 |
157
+ | 0.2636 | 3.41 | 98 | 0.5744 |
158
+ | 0.3114 | 3.67 | 105 | 0.5741 |
159
+
160
+
161
+ ### Framework versions
162
+
163
+ - PEFT 0.9.0
164
+ - Transformers 4.38.2
165
+ - Pytorch 2.1.2+cu121
166
+ - Datasets 2.18.0
167
+ - Tokenizers 0.15.0
adapter_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9165080527686729d26aaaf8d6a848a2dd5c613a2f51146d57b882e461def334
3
+ size 84047370