Hugging Face's logo

Weyaxi
/
Einstein-7B

Text Generation
Transformers
PyTorch
mistral
axolotl
Generated from Trainer
text-generation-inference
Inference Endpoints
Model card Files Community
2
Weyaxi commited on
Commit
36f6450
·
verified ·
1 Parent(s): 8c1fcc1

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +195 -0
README.md ADDED
@@ -0,0 +1,195 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ tags:
4
+ - axolotl
5
+ - generated_from_trainer
6
+ base_model: mistralai/Mistral-7B-v0.1
7
+ ---
8
+
9
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
10
+ should probably proofread and complete it, then remove this comment. -->
11
+
12
+ [<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)
13
+ <details><summary>See axolotl config</summary>
14
+
15
+ axolotl version: `0.3.0`
16
+ ```yaml
17
+ base_model: mistralai/Mistral-7B-v0.1
18
+ model_type: MistralForCausalLM
19
+ tokenizer_type: LlamaTokenizer
20
+ is_mistral_derived_model: true
21
+
22
+ load_in_8bit: false
23
+ load_in_4bit: true
24
+ strict: false
25
+
26
+ datasets:
27
+ - path: sci-datasets/arc_challange_train_alpaca.json
28
+ ds_type: json
29
+ type: alpaca
30
+
31
+ - path: sci-datasets/camelai_biology_alpaca.json
32
+ ds_type: json
33
+ type: alpaca
34
+
35
+ - path: sci-datasets/camelai_chemistry_alpaca.json
36
+ ds_type: json
37
+ type: alpaca
38
+
39
+ - path: sci-datasets/camelai_physics_alpaca.json
40
+ ds_type: json
41
+ type: alpaca
42
+
43
+ - path: sci-datasets/openbookqa_alpaca.json
44
+ ds_type: json
45
+ type: alpaca
46
+
47
+ - path: sci-datasets/reclor_science_alpaca.json
48
+ ds_type: json
49
+ type: alpaca
50
+
51
+ - path: sci-datasets/scibench_alpaca.json
52
+ ds_type: json
53
+ type: alpaca
54
+
55
+ - path: sci-datasets/scienceqa_alpaca.json
56
+ ds_type: json
57
+ type: alpaca
58
+
59
+ - path: sci-datasets/theoremqa_alpaca.json
60
+ ds_type: json
61
+ type: alpaca
62
+
63
+ - path: sci-datasets/tiger_scienceeval_alpaca.json
64
+ ds_type: json
65
+ type: alpaca
66
+
67
+ dataset_prepared_path: last_run_prepared
68
+ val_set_size: 0
69
+ output_dir: ./science-mistral
70
+
71
+ adapter: qlora
72
+ lora_model_dir:
73
+
74
+ sequence_len: 8192
75
+ sample_packing: true
76
+ pad_to_sequence_len: true
77
+
78
+ lora_r: 128
79
+ lora_alpha: 64
80
+ lora_dropout: 0.05
81
+ lora_target_linear: true
82
+ lora_fan_in_fan_out:
83
+ lora_target_modules:
84
+ - gate_proj
85
+ - down_proj
86
+ - up_proj
87
+ - q_proj
88
+ - v_proj
89
+ - k_proj
90
+ - o_proj
91
+
92
+ wandb_project: huggingface
93
+ wandb_entity:
94
+ wandb_watch:
95
+ wandb_name:
96
+ wandb_log_model:
97
+ hub_model_id: Weyaxi/science-mistral
98
+
99
+ # change #
100
+ gradient_accumulation_steps: 12
101
+ micro_batch_size: 6
102
+ num_epochs: 2
103
+ optimizer: adamw_bnb_8bit
104
+ lr_scheduler: cosine
105
+ learning_rate: 0.0002
106
+ # change #
107
+
108
+ train_on_inputs: false
109
+ group_by_length: false
110
+ bf16: true
111
+ fp16: false
112
+ tf32: false
113
+
114
+ gradient_checkpointing: true
115
+ early_stopping_patience:
116
+ resume_from_checkpoint:
117
+ local_rank:
118
+ logging_steps: 1
119
+ xformers_attention:
120
+ flash_attention: true
121
+
122
+ warmup_steps: 10
123
+
124
+
125
+ saves_per_epoch: 3
126
+ debug:
127
+ deepspeed:
128
+ weight_decay: 0.1
129
+ fsdp:
130
+ fsdp_config:
131
+ special_tokens:
132
+ bos_token: "<s>"
133
+ eos_token: "</s>"
134
+ unk_token: "<unk>"
135
+
136
+ ```
137
+
138
+ </details><br>
139
+
140
+ # science-mistral
141
+
142
+ This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the None dataset.
143
+
144
+ ## Model description
145
+
146
+ More information needed
147
+
148
+ ## Intended uses & limitations
149
+
150
+ More information needed
151
+
152
+ ## Training and evaluation data
153
+
154
+ More information needed
155
+
156
+ ## Training procedure
157
+
158
+
159
+ The following `bitsandbytes` quantization config was used during training:
160
+ - quant_method: bitsandbytes
161
+ - load_in_8bit: False
162
+ - load_in_4bit: True
163
+ - llm_int8_threshold: 6.0
164
+ - llm_int8_skip_modules: None
165
+ - llm_int8_enable_fp32_cpu_offload: False
166
+ - llm_int8_has_fp16_weight: False
167
+ - bnb_4bit_quant_type: nf4
168
+ - bnb_4bit_use_double_quant: True
169
+ - bnb_4bit_compute_dtype: bfloat16
170
+
171
+ ### Training hyperparameters
172
+
173
+ The following hyperparameters were used during training:
174
+ - learning_rate: 0.0002
175
+ - train_batch_size: 6
176
+ - eval_batch_size: 6
177
+ - seed: 42
178
+ - gradient_accumulation_steps: 12
179
+ - total_train_batch_size: 72
180
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
181
+ - lr_scheduler_type: cosine
182
+ - lr_scheduler_warmup_steps: 10
183
+ - num_epochs: 2
184
+
185
+ ### Training results
186
+
187
+
188
+
189
+ ### Framework versions
190
+
191
+ - PEFT 0.7.0
192
+ - Transformers 4.37.0
193
+ - Pytorch 2.0.1+cu118
194
+ - Datasets 2.16.1
195
+ - Tokenizers 0.15.0