Weyaxi commited on
Commit
e7bed8c
1 Parent(s): 1096b82

End of training

Browse files
Files changed (1) hide show
  1. README.md +178 -0
README.md ADDED
@@ -0,0 +1,178 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ library_name: peft
4
+ tags:
5
+ - axolotl
6
+ - generated_from_trainer
7
+ base_model: Qwen/Qwen1.5-32B
8
+ model-index:
9
+ - name: Einstein-v4-Qwen-1.5-32B
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: Qwen/Qwen1.5-32B
22
+ model_type: AutoModelForCausalLM
23
+ tokenizer_type: AutoTokenizer
24
+
25
+ load_in_8bit: false
26
+ load_in_4bit: true
27
+ strict: false
28
+
29
+ chat_template: chatml
30
+ datasets:
31
+ - path: data/merged_all.json
32
+ ds_type: json
33
+ type: alpaca
34
+ conversation: chatml
35
+
36
+ - path: data/capybara_sharegpt.json
37
+ ds_type: json
38
+ type: sharegpt
39
+ conversation: chatml
40
+
41
+ - path: data/synthia-v1.3_sharegpt_12500.json
42
+ ds_type: json
43
+ type: sharegpt
44
+ conversation: chatml
45
+
46
+ - path: data/cot_alpaca_gpt4_extracted_openhermes_2.5_sharegpt.json
47
+ ds_type: json
48
+ type: sharegpt
49
+ conversation: chatml
50
+
51
+ - path: data/slimorca_dedup_filtered_95k_sharegpt.json
52
+ ds_type: json
53
+ type: sharegpt
54
+ conversation: chatml
55
+
56
+ - path: data/airoboros_3.2_without_contextual_slimorca_orca_sharegpt.json
57
+ ds_type: json
58
+ type: sharegpt
59
+ conversation: chatml
60
+
61
+ dataset_prepared_path: last_run_prepared
62
+ val_set_size: 0 # because we won't eval, out of memory :(
63
+ output_dir: ./Einstein-v4-Qwen-1.5-32B-model
64
+
65
+ sequence_len: 4096
66
+ sample_packing: true
67
+ pad_to_sequence_len: true
68
+ eval_sample_packing: false
69
+
70
+ adapter: qlora
71
+ lora_model_dir:
72
+ lora_r: 64
73
+ lora_alpha: 32
74
+ lora_dropout: 0.05
75
+ lora_target_linear: true
76
+ lora_fan_in_fan_out:
77
+ lora_modules_to_save:
78
+ - "embed_tokens"
79
+ - "lm_head"
80
+
81
+ wandb_project: Einstein
82
+ wandb_entity:
83
+ wandb_watch:
84
+ wandb_name: Einstein-v4-Qwen-1.5-32B-qlora-2-epoch
85
+ wandb_log_model:
86
+ hub_model_id: Weyaxi/Einstein-v4-Qwen-1.5-32B
87
+
88
+ save_safetensors: true
89
+
90
+ gradient_accumulation_steps: 4
91
+ micro_batch_size: 1
92
+ num_epochs: 2
93
+ optimizer: adamw_bnb_8bit
94
+ lr_scheduler: cosine
95
+ learning_rate: 0.0002
96
+
97
+ train_on_inputs: false
98
+ group_by_length: false
99
+ bf16: true
100
+ fp16: false
101
+ tf32: false
102
+
103
+ gradient_checkpointing: true
104
+ early_stopping_patience:
105
+ resume_from_checkpoint:
106
+ local_rank:
107
+ logging_steps: 1
108
+ xformers_attention:
109
+ flash_attention: true
110
+
111
+ warmup_steps: 10
112
+ evals_per_epoch: 0 # because we won't eval, out of memory :(
113
+ eval_table_size:
114
+ eval_table_max_new_tokens: 128
115
+ saves_per_epoch: 2
116
+ debug:
117
+
118
+ deepspeed: zero3_bf16_cpuoffload_params.json
119
+ weight_decay: 0.0
120
+ fsdp:
121
+ fsdp_config:
122
+ special_tokens:
123
+ bos_token: "<s>"
124
+ eos_token: "<|im_end|>"
125
+ unk_token: "<unk>"
126
+ tokens:
127
+ - "<|im_start|>"
128
+
129
+ ```
130
+
131
+ </details><br>
132
+
133
+ # Einstein-v4-Qwen-1.5-32B
134
+
135
+ This model is a fine-tuned version of [Qwen/Qwen1.5-32B](https://huggingface.co/Qwen/Qwen1.5-32B) on the None dataset.
136
+
137
+ ## Model description
138
+
139
+ More information needed
140
+
141
+ ## Intended uses & limitations
142
+
143
+ More information needed
144
+
145
+ ## Training and evaluation data
146
+
147
+ More information needed
148
+
149
+ ## Training procedure
150
+
151
+ ### Training hyperparameters
152
+
153
+ The following hyperparameters were used during training:
154
+ - learning_rate: 0.0002
155
+ - train_batch_size: 1
156
+ - eval_batch_size: 1
157
+ - seed: 42
158
+ - distributed_type: multi-GPU
159
+ - num_devices: 9
160
+ - gradient_accumulation_steps: 4
161
+ - total_train_batch_size: 36
162
+ - total_eval_batch_size: 9
163
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
164
+ - lr_scheduler_type: cosine
165
+ - lr_scheduler_warmup_steps: 10
166
+ - num_epochs: 2
167
+
168
+ ### Training results
169
+
170
+
171
+
172
+ ### Framework versions
173
+
174
+ - PEFT 0.10.0
175
+ - Transformers 4.40.0.dev0
176
+ - Pytorch 2.1.2+cu118
177
+ - Datasets 2.18.0
178
+ - Tokenizers 0.15.0