File size: 6,006 Bytes
fc0acaf
 
 
 
 
 
 
 
 
 
 
 
 
b681162
fc0acaf
 
 
 
e08015b
fc0acaf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
59e2f44
fc0acaf
59e2f44
fc0acaf
 
dfc3911
fc0acaf
 
 
 
7b63113
 
 
 
 
 
 
 
 
 
 
 
132915b
7b63113
 
 
 
 
 
 
 
 
 
 
 
dfc3911
 
 
 
 
 
 
 
 
 
 
 
7b63113
fc0acaf
 
 
 
 
 
 
 
 
 
 
 
b681162
 
 
 
 
 
 
 
fc0acaf
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
---
license: mit
datasets:
- TIGER-Lab/SKGInstruct
language:
- en
---
# 🏗️ StructLM: Towards Building Generalist Models for Structured Knowledge Grounding



Project Page: [https://tiger-ai-lab.github.io/StructLM/](https://tiger-ai-lab.github.io/StructLM/)

Paper: [https://arxiv.org/pdf/2402.16671.pdf](https://arxiv.org/pdf/2402.16671.pdf)

Code: [https://github.com/TIGER-AI-Lab/StructLM](https://github.com/TIGER-AI-Lab/StructLM)


![Alt text](https://raw.githubusercontent.com/TIGER-AI-Lab/StructLM/gh-pages/static/images/thumbnail.drawio.png)


## Introduction
StructLM, is a series of open-source large language models (LLMs) finetuned for structured knowledge grounding (SKG) tasks. We release 3 models:

 7B  | [StructLM-7B](https://huggingface.co/TIGER-Lab/StructLM-7B)   
 
 13B | [StructLM-13B](https://huggingface.co/TIGER-Lab/StructLM-13B) 
 
 34B | [StructLM-34B](https://huggingface.co/TIGER-Lab/StructLM-34B) 


## Training Data
These models are trained on 🤗 [SKGInstruct Dataset](https://huggingface.co/datasets/TIGER-Lab/SKGInstruct), an instruction-tuning dataset containing mixture of 19 SKG tasks combined with 🤗 [SlimOrca](https://huggingface.co/datasets/Open-Orca/SlimOrca).  Check out the dataset card for more details.


## Training Procedure
The models are fine-tuned with CodeLlama-Instruct-hf models as base models. Each model is trained for 3 epochs, and the best checkpoint is selected.

## Evaluation
Here are a subset of model evaluation results:

### Held in

| **Model**             	| **ToTTo** 	| **GrailQA**  	| **CompWebQ** 	| **MMQA** 	| **Feverous** 	| **Spider**  	| **TabFact**  	| **Dart**  	|
|-----------------------|--------------|----------|----------|----------|----------|----------|----------|----------|
| **StructLM-7B**        	| 49.4          	| 80.4     	| 78.3  	| 85.2     	| 84.4     	| 72.4     	| 80.8      	| 62.2     	|
| **StructLM-13B**  	| 49.3          	| 79.2    	| 80.4  	| 86.0     	| 85.0     	| 74.1     	| 84.7     	| 61.4      	| 
| **StructLM-34B**       	| 50.2          	| 82.2     	| 81.9 	| 88.1     	| 85.7     	| 74.6     	| 86.6     	| 61.8     	| 


### Held out
| **Model**             	| **BIRD** 	| **InfoTabs**  	| **FinQA** 	| **SQA** 	|
|-----------------------|--------------|----------|----------|----------|
| **StructLM-7B**        	| 22.3          	| 55.3     	| 27.3     	| 49.7     	|
| **StructLM-13B**  	| 22.8          	| 58.1     	| 25.6      	| 36.1     	| 
| **StructLM-34B**       	| 24.7          	| 61.8     	| 36.2     	| 44.2     	| 


## Usage
You can use the models through Huggingface's Transformers library. 
Check our Github repo for the evaluation code: [https://github.com/TIGER-AI-Lab/StructLM](https://github.com/TIGER-AI-Lab/StructLM)


## Prompt Format

\*\*\***IMPORTANT**\*\*\*

**For this 13B model, the prompt format (different from 7B) is**
```
[INST] [INST] <<SYS>>
You are an AI assistant that specializes in analyzing and reasoning over structured information. You will be given a task, optionally with some structured knowledge input. Your answer must strictly adhere to the output format, if specified.
<</SYS>>
{instruction} [/INST] [/INST]
```


To see concrete examples of this linearization, you can directly reference the 🤗 [SKGInstruct Dataset](https://huggingface.co/datasets/TIGER-Lab/SKGInstruct) (coming soon).
We will provide code for linearizing this data shortly.


A few examples:

**tabular data**
```
col : day | kilometers row 1 : tuesday | 0 row 2 : wednesday | 0 row 3 : thursday | 4 row 4 : friday | 0 row 5 : saturday | 0
```

**knowledge triples (dart)**
```
Hawaii Five-O : notes : Episode: The Flight of the Jewels | [TABLECONTEXT] : [title] : Jeff Daniels | [TABLECONTEXT] : title : Hawaii Five-O
```

**knowledge graph schema (grailqa)**
```
top antiquark: m.094nrqp | physics.particle_antiparticle.self_antiparticle physics.particle_family physics.particle.antiparticle physics.particle_family.subclasses physics.subatomic_particle_generation physics.particle_family.particles physics.particle common.image.appears_in_topic_gallery physics.subatomic_particle_generation.particles physics.particle.family physics.particle_family.parent_class physics.particle_antiparticle physics.particle_antiparticle.particle physics.particle.generation
```

**example input**

```
[INST] [INST] <<SYS>>
You are an AI assistant that specializes in analyzing and reasoning over structured information. You will be given a task, optionally with some structured knowledge input. Your answer must strictly adhere to the output format, if specified.
<</SYS>>

Use the information in the following table to solve the problem, choose between the choices if they are provided. table:

col : day | kilometers row 1 : tuesday | 0 row 2 : wednesday | 0 row 3 : thursday | 4 row 4 : friday | 0 row 5 : saturday | 0


question:

Allie kept track of how many kilometers she walked during the past 5 days. What is the range of the numbers? [/INST] [/INST]
```

## Intended Uses
These models are trained for research purposes. They are designed to be proficient in interpreting linearized structured input. Downstream uses can potentially include various applications requiring the interpretation of structured data.

## Limitations
While we've tried to build an SKG-specialized model capable of generalizing, we have shown that this is a challenging domain, and it may lack performance characteristics that allow it to be directly used in chat or other applications.


## Citation
If you use the models, data, or code from this project, please cite the original paper:

```
@misc{zhuang2024structlm,
      title={StructLM: Towards Building Generalist Models for Structured Knowledge Grounding}, 
      author={Alex Zhuang and Ge Zhang and Tianyu Zheng and Xinrun Du and Junjie Wang and Weiming Ren and Stephen W. Huang and Jie Fu and Xiang Yue and Wenhu Chen},
      year={2024},
      eprint={2402.16671},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```