File size: 2,213 Bytes
78e1c06
 
 
 
 
 
 
 
 
 
 
 
f16a0a4
78e1c06
 
 
 
9503510
78e1c06
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f16a0a4
78e1c06
2130e11
78e1c06
 
2130e11
 
 
 
 
 
 
 
78e1c06
 
 
 
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
---
language:
- en
datasets:
- tatsu-lab/alpaca
---
# Model Card for Model ID

This model checkpoint is the TinyLlama-1.1B fine-tuned on [alpaca dataset](https://huggingface.co/datasets/tatsu-lab/alpaca).

## Model Details

### Model Sources 

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

- **Repository:** https://github.com/jzhang38/TinyLlama
- **Paper:** [https://arxiv.org/abs/2404.02406]

## Uses


The use of this model should comply with the restrictions from [TinyLlama-1.1b](https://github.com/jzhang38/TinyLlama) and 
[Stanford Alpaca](https://github.com/tatsu-lab/stanford_alpaca).

## How to Get Started with the Model

Use the code below to get started with the model.

```
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("luckychao/TinyAlpaca-1.1B")
model = AutoModelForCausalLM.from_pretrained("luckychao/TinyAlpaca-1.1B")

```

## Training Details

### Training Data

We use the [alpaca dataset](https://huggingface.co/datasets/tatsu-lab/alpaca), which is created by researchers from Stanford University.

### Training Procedure

We follow the same training procedure and mostly same hyper-parameters to fine-tune the original Alpaca model on Llama. The procedure can be found in [stanford_alpaca project](https://huggingface.co/datasets/tatsu-lab/alpaca).

#### Training Hyperparameters
```
--num_train_epochs 3 \
--per_device_train_batch_size 2 \
--per_device_eval_batch_size 2 \
--gradient_accumulation_steps 4 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 1000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--bf16 True \
--fsdp "full_shard auto_wrap" \
--fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
--model_max_length 2048 

```


## Citation 

The model is mostly developed for the paper below. Please cite it if you find the repository helpful.

**BibTeX:**
```
@article{hao2024exploring,
  title={Exploring Backdoor Vulnerabilities of Chat Models},
  author={Hao, Yunzhuo and Yang, Wenkai and Lin, Yankai},
  journal={arXiv preprint arXiv:2404.02406},
  year={2024}
}
```