File size: 5,286 Bytes
29c8585
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd6900e
29c8585
 
 
cd6900e
29c8585
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cd6900e
 
 
 
 
 
 
 
29c8585
 
 
 
 
 
 
 
 
cd6900e
 
 
 
29c8585
 
 
 
 
 
cd6900e
 
29c8585
 
 
 
 
 
cd6900e
 
29c8585
 
 
 
 
 
cd6900e
29c8585
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
---
language:
- en
pipeline_tag: zero-shot-classification
tags:
- finance
- compliance
---

# Model Card for Model ID

<!-- 
 -->

## Model Details
Based of the full weight llama 2-hermes from Nous Research. 

### Model Description
This model was fine tuned off the full weight llama-2-hermes-7B from Nous Research. This model is a preemptive V1, and a hastily put together model to assist 
in finance and compliance tasks, mostly tuned to the new SEC Marketing and Compliance rules established in 2021. Later iterations will have more guidelines and rulings 
unrelated to the SEC Marketing rule. 
https://www.sec.gov/files/rules/final/2020/ia-5653.pdf
<!-- -->



- **Developed by:** [More Information Needed]
- **Shared by [optional]:** [More Information Needed]
- **Model type:** [More Information Needed]
- **Language(s) (NLP):** [Enlgish]
- **License:** [More Information Needed]
- **Finetuned from model [optional]:** [llama 2-hermes-7b]

### Model Sources [optional]

<!-- Provide the basic links for the model. -->

- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]

## Uses

This is to help companies and individuals within compliance and marketing departments to determine and find issues within their marketing or public facing documents.
Since the new marketing rule is principles based it requires logic, experience, and reasoning to determine if a statement or advertisement would be compliant within 
the SEC's new guidelines. This can lead to multiple viewpoints of compliant or not depending on the viewer. Thus this is a small/high quality dataset version 
to aid or provide an second viewpoint of a public facing statement to help determine if something is compliant per the SEC's guidelines. The dataset was crafted by
reviewing the SEC Marketing rule, other scenarios, and providing reasoning within the ###n\ Response n\### to help guide the model in reasoning tasks. 
Further versions will be reviewed more for accuracy, bias, and more data. 

<!-- -->

### Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

[More Information Needed]

### Downstream Use [optional]

For use by marketing and compliance finance teams to assist in determination and interpretation of SEC Marketing rule and other SEC interpretations. No outputs should be guaranteed as fact,
and review of data is encouraged. This is to simply assist, and aid those in remembering certain aspects and interpretation of aspects of the long SEC Marketing guidelines 
amongst other SEC rulings.

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

[More Information Needed]

### Out-of-Scope Use

This model should not be intended to be used as fact, as evidence/proof in a trial hearing, or be used as indication of innocence in an SEC audit/investigation. 
This model should be used by professionals deeply familiar with the SEC's guidelines and compliance procedures. 
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

[More Information Needed]

## Bias, Risks, and Limitations

This is the first model iteration, and has not be fully reviewed by multiple professional peers for its accuracy, bias, and output variations.  
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model.
<!-- This section is meant to convey both technical and sociotechnical limitations. -->

[More Information Needed]

### Recommendations

<!-- Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. --> 

## How to Get Started with the Model

Use the code below to get started with the model.

[More Information Needed]

## Training Details

### Training Data

<!--  -->


### Training Procedure 

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->


#### Training Hyperparameters

- <!--# Compute dtype for 4-bit base models
bnb_4bit_compute_dtype = "float16"

bnb_4bit_quant_type = "nf4"
use_nested_quant = False
fp16 = False
bf16 = False - this will be True for next training run. 
per_device_train_batch_size = 4
per_device_eval_batch_size = 4
gradient_accumulation_steps = 1
gradient_checkpointing = True
max_grad_norm = 0.3
learning_rate = 2e-5 -1 e-4 for a 13B will be applied. 
weight_decay = 0.001
optim = "paged_adamw_32bit"
lr_scheduler_type = "constant"
max_steps = 13000
warmup_ratio = 0.03
group_by_length = True
 -->



## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Data Card if possible. -->



[More Information Needed]

#### Metrics

<!--  -->

[More Information Needed]

### Results

[More Information Needed]

#### Summary



## Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

### Model Architecture and Objective

[More Information Needed]

### Compute Infrastructure

[Google Colab]

#### Hardware

[1xA100]