Files changed (1) hide show
  1. README.md +168 -154
README.md CHANGED
@@ -1,155 +1,169 @@
1
- ---
2
- library_name: peft
3
- license: apache-2.0
4
- base_model: Qwen/Qwen2.5-1.5B
5
- tags:
6
- - axolotl
7
- - generated_from_trainer
8
- model-index:
9
- - name: 3df69a64-ceb5-4f98-aa56-24a10bee36e0
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/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
17
- <details><summary>See axolotl config</summary>
18
-
19
- axolotl version: `0.4.1`
20
- ```yaml
21
- adapter: lora
22
- base_model: Qwen/Qwen2.5-1.5B
23
- bf16: auto
24
- chat_template: llama3
25
- dataset_prepared_path: null
26
- datasets:
27
- - data_files:
28
- - 30467321b0218499_train_data.json
29
- ds_type: json
30
- format: custom
31
- path: /workspace/input_data/30467321b0218499_train_data.json
32
- type:
33
- field_input: Japanese
34
- field_instruction: ENName
35
- field_output: English
36
- format: '{instruction} {input}'
37
- no_input_format: '{instruction}'
38
- system_format: '{system}'
39
- system_prompt: ''
40
- debug: null
41
- deepspeed: null
42
- device_map: auto
43
- do_eval: true
44
- early_stopping_patience: null
45
- eval_batch_size: 4
46
- eval_max_new_tokens: 128
47
- eval_steps: null
48
- eval_table_size: null
49
- evals_per_epoch: null
50
- flash_attention: true
51
- fp16: null
52
- fsdp: null
53
- fsdp_config: null
54
- gradient_accumulation_steps: 4
55
- gradient_checkpointing: true
56
- group_by_length: true
57
- hub_model_id: ancient41/3df69a64-ceb5-4f98-aa56-24a10bee36e0
58
- hub_repo: null
59
- hub_strategy: end
60
- hub_token: null
61
- learning_rate: 0.0001
62
- load_in_4bit: false
63
- load_in_8bit: false
64
- local_rank: null
65
- logging_steps: 1
66
- lora_alpha: 16
67
- lora_dropout: 0.05
68
- lora_fan_in_fan_out: null
69
- lora_model_dir: null
70
- lora_r: 8
71
- lora_target_linear: true
72
- lr_scheduler: cosine
73
- max_grad_norm: 1.0
74
- max_steps: 1256
75
- micro_batch_size: 4
76
- mlflow_experiment_name: /tmp/30467321b0218499_train_data.json
77
- model_type: AutoModelForCausalLM
78
- num_epochs: 1
79
- optimizer: adamw_bnb_8bit
80
- output_dir: miner_id_24
81
- pad_to_sequence_len: true
82
- resume_from_checkpoint: null
83
- s2_attention: null
84
- sample_packing: false
85
- save_steps: null
86
- saves_per_epoch: 4
87
- sequence_len: 1024
88
- strict: false
89
- tf32: true
90
- tokenizer_type: AutoTokenizer
91
- train_on_inputs: false
92
- trust_remote_code: true
93
- val_set_size: 0.05
94
- wandb_entity: null
95
- wandb_mode: online
96
- wandb_name: f01f50dd-806e-4a8f-9c25-300f4646bd3d
97
- wandb_project: Gradients-On-Demand
98
- wandb_run: your_name
99
- wandb_runid: f01f50dd-806e-4a8f-9c25-300f4646bd3d
100
- warmup_steps: 5
101
- weight_decay: 0.0
102
- xformers_attention: null
103
-
104
- ```
105
-
106
- </details><br>
107
-
108
- # 3df69a64-ceb5-4f98-aa56-24a10bee36e0
109
-
110
- This model is a fine-tuned version of [Qwen/Qwen2.5-1.5B](https://huggingface.co/Qwen/Qwen2.5-1.5B) on the None dataset.
111
- It achieves the following results on the evaluation set:
112
- - Loss: 1.4277
113
-
114
- ## Model description
115
-
116
- More information needed
117
-
118
- ## Intended uses & limitations
119
-
120
- More information needed
121
-
122
- ## Training and evaluation data
123
-
124
- More information needed
125
-
126
- ## Training procedure
127
-
128
- ### Training hyperparameters
129
-
130
- The following hyperparameters were used during training:
131
- - learning_rate: 0.0001
132
- - train_batch_size: 4
133
- - eval_batch_size: 4
134
- - seed: 42
135
- - gradient_accumulation_steps: 4
136
- - total_train_batch_size: 16
137
- - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
138
- - lr_scheduler_type: cosine
139
- - lr_scheduler_warmup_steps: 5
140
- - training_steps: 1256
141
-
142
- ### Training results
143
-
144
- | Training Loss | Epoch | Step | Validation Loss |
145
- |:-------------:|:------:|:----:|:---------------:|
146
- | 1.4374 | 0.6233 | 1256 | 1.4277 |
147
-
148
-
149
- ### Framework versions
150
-
151
- - PEFT 0.13.2
152
- - Transformers 4.46.0
153
- - Pytorch 2.5.0+cu124
154
- - Datasets 3.0.1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
155
  - Tokenizers 0.20.1
 
1
+ ---
2
+ library_name: peft
3
+ license: apache-2.0
4
+ base_model: Qwen/Qwen2.5-1.5B
5
+ tags:
6
+ - axolotl
7
+ - generated_from_trainer
8
+ language:
9
+ - zho
10
+ - eng
11
+ - fra
12
+ - spa
13
+ - por
14
+ - deu
15
+ - ita
16
+ - rus
17
+ - jpn
18
+ - kor
19
+ - vie
20
+ - tha
21
+ - ara
22
+ model-index:
23
+ - name: 3df69a64-ceb5-4f98-aa56-24a10bee36e0
24
+ results: []
25
+ ---
26
+
27
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
28
+ should probably proofread and complete it, then remove this comment. -->
29
+
30
+ [<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
31
+ <details><summary>See axolotl config</summary>
32
+
33
+ axolotl version: `0.4.1`
34
+ ```yaml
35
+ adapter: lora
36
+ base_model: Qwen/Qwen2.5-1.5B
37
+ bf16: auto
38
+ chat_template: llama3
39
+ dataset_prepared_path: null
40
+ datasets:
41
+ - data_files:
42
+ - 30467321b0218499_train_data.json
43
+ ds_type: json
44
+ format: custom
45
+ path: /workspace/input_data/30467321b0218499_train_data.json
46
+ type:
47
+ field_input: Japanese
48
+ field_instruction: ENName
49
+ field_output: English
50
+ format: '{instruction} {input}'
51
+ no_input_format: '{instruction}'
52
+ system_format: '{system}'
53
+ system_prompt: ''
54
+ debug: null
55
+ deepspeed: null
56
+ device_map: auto
57
+ do_eval: true
58
+ early_stopping_patience: null
59
+ eval_batch_size: 4
60
+ eval_max_new_tokens: 128
61
+ eval_steps: null
62
+ eval_table_size: null
63
+ evals_per_epoch: null
64
+ flash_attention: true
65
+ fp16: null
66
+ fsdp: null
67
+ fsdp_config: null
68
+ gradient_accumulation_steps: 4
69
+ gradient_checkpointing: true
70
+ group_by_length: true
71
+ hub_model_id: ancient41/3df69a64-ceb5-4f98-aa56-24a10bee36e0
72
+ hub_repo: null
73
+ hub_strategy: end
74
+ hub_token: null
75
+ learning_rate: 0.0001
76
+ load_in_4bit: false
77
+ load_in_8bit: false
78
+ local_rank: null
79
+ logging_steps: 1
80
+ lora_alpha: 16
81
+ lora_dropout: 0.05
82
+ lora_fan_in_fan_out: null
83
+ lora_model_dir: null
84
+ lora_r: 8
85
+ lora_target_linear: true
86
+ lr_scheduler: cosine
87
+ max_grad_norm: 1.0
88
+ max_steps: 1256
89
+ micro_batch_size: 4
90
+ mlflow_experiment_name: /tmp/30467321b0218499_train_data.json
91
+ model_type: AutoModelForCausalLM
92
+ num_epochs: 1
93
+ optimizer: adamw_bnb_8bit
94
+ output_dir: miner_id_24
95
+ pad_to_sequence_len: true
96
+ resume_from_checkpoint: null
97
+ s2_attention: null
98
+ sample_packing: false
99
+ save_steps: null
100
+ saves_per_epoch: 4
101
+ sequence_len: 1024
102
+ strict: false
103
+ tf32: true
104
+ tokenizer_type: AutoTokenizer
105
+ train_on_inputs: false
106
+ trust_remote_code: true
107
+ val_set_size: 0.05
108
+ wandb_entity: null
109
+ wandb_mode: online
110
+ wandb_name: f01f50dd-806e-4a8f-9c25-300f4646bd3d
111
+ wandb_project: Gradients-On-Demand
112
+ wandb_run: your_name
113
+ wandb_runid: f01f50dd-806e-4a8f-9c25-300f4646bd3d
114
+ warmup_steps: 5
115
+ weight_decay: 0.0
116
+ xformers_attention: null
117
+
118
+ ```
119
+
120
+ </details><br>
121
+
122
+ # 3df69a64-ceb5-4f98-aa56-24a10bee36e0
123
+
124
+ This model is a fine-tuned version of [Qwen/Qwen2.5-1.5B](https://huggingface.co/Qwen/Qwen2.5-1.5B) on the None dataset.
125
+ It achieves the following results on the evaluation set:
126
+ - Loss: 1.4277
127
+
128
+ ## Model description
129
+
130
+ More information needed
131
+
132
+ ## Intended uses & limitations
133
+
134
+ More information needed
135
+
136
+ ## Training and evaluation data
137
+
138
+ More information needed
139
+
140
+ ## Training procedure
141
+
142
+ ### Training hyperparameters
143
+
144
+ The following hyperparameters were used during training:
145
+ - learning_rate: 0.0001
146
+ - train_batch_size: 4
147
+ - eval_batch_size: 4
148
+ - seed: 42
149
+ - gradient_accumulation_steps: 4
150
+ - total_train_batch_size: 16
151
+ - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
152
+ - lr_scheduler_type: cosine
153
+ - lr_scheduler_warmup_steps: 5
154
+ - training_steps: 1256
155
+
156
+ ### Training results
157
+
158
+ | Training Loss | Epoch | Step | Validation Loss |
159
+ |:-------------:|:------:|:----:|:---------------:|
160
+ | 1.4374 | 0.6233 | 1256 | 1.4277 |
161
+
162
+
163
+ ### Framework versions
164
+
165
+ - PEFT 0.13.2
166
+ - Transformers 4.46.0
167
+ - Pytorch 2.5.0+cu124
168
+ - Datasets 3.0.1
169
  - Tokenizers 0.20.1