File size: 5,287 Bytes
9530d8e
 
 
 
 
3c5a32a
9530d8e
 
 
3c5a32a
 
9530d8e
 
 
3c5a32a
9530d8e
3c5a32a
9530d8e
 
 
 
 
3c5a32a
9530d8e
 
 
3c5a32a
 
 
9530d8e
 
 
3c5a32a
 
 
 
 
 
 
 
 
 
 
9530d8e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3c5a32a
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
---
license: apache-2.0
base_model: mistralai/Mistral-7B-v0.1
tags:
- generated_from_trainer
- not-for-all-audiences
model-index:
- name: pippa-lora
  results: []
datasets:
- PygmalionAI/PIPPA
---

[<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)
# mistral-v0.1-7b-pippa-metharme-lora

This model is a fine-tuned version of [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1) on the PIPPA dataset.
It achieves the following results on the evaluation set:
- Loss: 1.3494

## Model description

8-bit lora trained on the [PygmalionAI/PIPPA](https://huggingface.co/datasets/PygmalionAI/PIPPA) dataset using axolotl.

## Intended uses & limitations

PIPPA consists of just a little more than 1 million lines of dialogue spread out over 26,000 conversations between users of the popular chatbot website "Character.AI" and its large language model, obtained through a large community effort taking place over the course of several months. Tallying shows that over 1,000 unique personas simulating both real and fictional characters are represented within the dataset, allowing PIPPA and LLMs fine-tuned on it to adapt to many different roleplay domains.

⚠️ CAUTION: PIPPA contains conversations, themes and scenarios which can be considered "not safe for work" (NSFW) and/or heavily disturbing in nature. Models trained purely with PIPPA may have the tendency to generate X-rated output. You have been warned.

## Training and evaluation data

[PygmalionAI/PIPPA](https://huggingface.co/datasets/PygmalionAI/PIPPA)
```
@misc{gosling2023pippa,
      title={PIPPA: A Partially Synthetic Conversational Dataset}, 
      author={Tear Gosling and Alpin Dale and Yinhe Zheng},
      year={2023},
      eprint={2308.05884},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 2
- eval_batch_size: 2
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 40
- num_epochs: 3

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 1.7313        | 0.05  | 100  | 1.7044          |
| 1.68          | 0.11  | 200  | 1.6176          |
| 1.5642        | 0.16  | 300  | 1.5538          |
| 1.6617        | 0.22  | 400  | 1.4986          |
| 1.4733        | 0.27  | 500  | 1.4723          |
| 1.4916        | 0.33  | 600  | 1.4427          |
| 1.5036        | 0.38  | 700  | 1.4271          |
| 1.2385        | 0.44  | 800  | 1.4109          |
| 1.4094        | 0.49  | 900  | 1.3968          |
| 1.4042        | 0.55  | 1000 | 1.3848          |
| 1.3946        | 0.6   | 1100 | 1.3771          |
| 1.2523        | 0.66  | 1200 | 1.3692          |
| 1.2932        | 0.71  | 1300 | 1.3648          |
| 1.346         | 0.77  | 1400 | 1.3609          |
| 1.1163        | 0.82  | 1500 | 1.3565          |
| 1.4656        | 0.88  | 1600 | 1.3495          |
| 1.2698        | 0.93  | 1700 | 1.3484          |
| 1.2019        | 0.99  | 1800 | 1.3454          |
| 1.3685        | 1.04  | 1900 | 1.3477          |
| 1.2248        | 1.1   | 2000 | 1.3488          |
| 1.2162        | 1.15  | 2100 | 1.3479          |
| 1.0443        | 1.21  | 2200 | 1.3491          |
| 1.2445        | 1.26  | 2300 | 1.3460          |
| 1.3229        | 1.32  | 2400 | 1.3476          |
| 1.3464        | 1.37  | 2500 | 1.3439          |
| 1.2651        | 1.43  | 2600 | 1.3439          |
| 1.516         | 1.48  | 2700 | 1.3424          |
| 1.4323        | 1.54  | 2800 | 1.3413          |
| 1.08          | 1.59  | 2900 | 1.3436          |
| 1.289         | 1.64  | 3000 | 1.3379          |
| 1.1221        | 1.7   | 3100 | 1.3384          |
| 1.1895        | 1.75  | 3200 | 1.3376          |
| 1.3138        | 1.81  | 3300 | 1.3358          |
| 1.3907        | 1.86  | 3400 | 1.3343          |
| 1.4544        | 1.92  | 3500 | 1.3351          |
| 1.25          | 1.97  | 3600 | 1.3334          |
| 1.2682        | 2.03  | 3700 | 1.3452          |
| 1.3107        | 2.08  | 3800 | 1.3471          |
| 1.2096        | 2.14  | 3900 | 1.3496          |
| 1.4503        | 2.19  | 4000 | 1.3503          |
| 1.142         | 2.25  | 4100 | 1.3485          |
| 0.8439        | 2.3   | 4200 | 1.3490          |
| 1.2749        | 2.36  | 4300 | 1.3508          |
| 0.9578        | 2.41  | 4400 | 1.3502          |
| 1.2203        | 2.47  | 4500 | 1.3496          |
| 0.9451        | 2.52  | 4600 | 1.3498          |
| 0.9602        | 2.58  | 4700 | 1.3491          |
| 0.9501        | 2.63  | 4800 | 1.3491          |
| 1.2062        | 2.69  | 4900 | 1.3496          |
| 1.1728        | 2.74  | 5000 | 1.3491          |
| 1.2506        | 2.8   | 5100 | 1.3494          |
| 1.4052        | 2.85  | 5200 | 1.3494          |
| 1.2012        | 2.91  | 5300 | 1.3494          |
| 1.3141        | 2.96  | 5400 | 1.3494          |


### Framework versions

- Transformers 4.34.0.dev0
- Pytorch 2.0.1+cu118
- Datasets 2.14.5
- Tokenizers 0.14.0