File size: 13,309 Bytes
4e96e0b
 
 
 
 
 
 
 
 
 
9f78c04
 
19c1068
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7f682b5
 
 
c790f55
b28447d
 
 
 
 
 
 
e94e001
b28447d
 
 
9f78c04
b28447d
823ff56
b28447d
6e73657
793a2fb
 
 
 
2054688
 
823ff56
793a2fb
 
8775347
793a2fb
823ff56
793a2fb
 
 
 
2054688
793a2fb
 
 
 
 
b28447d
2054688
b28447d
793a2fb
4ea418f
793a2fb
 
 
 
 
 
 
 
4ea418f
793a2fb
4ea418f
793a2fb
b28447d
 
4ea418f
793a2fb
 
 
 
 
b28447d
83bf16e
 
 
 
 
 
 
ec0d50b
 
48937fe
83bf16e
 
48937fe
83bf16e
 
d8fddf7
8ccc690
9607fcc
 
605c8a8
9607fcc
ca8d815
9607fcc
 
 
 
 
 
 
 
 
 
 
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
---
license: mit
datasets:
- disi-unibo-nlp/medqa-5-opt-MedGENIE
language:
- en
metrics:
- accuracy
tags:
- medical
- question-answering
- fusion-in-decoder
pipeline_tag: question-answering
widget:
- text: >-
    A junior orthopaedic surgery resident is completing a carpal tunnel repair
    with the department chairman as the attending physician. During the case,
    the resident inadvertently cuts a flexor tendon. The tendon is repaired
    without complication. The attending tells the resident that the patient will
    do fine, and there is no need to report this minor complication that will
    not harm the patient, as he does not want to make the patient worry
    unnecessarily. He tells the resident to leave this complication out of the
    operative report. Which of the following is the correct next action for the
    resident to take? A. Disclose the error to the patient and put it in the
    operative report B. Tell the attending that he cannot fail to disclose this
    mistake C. Report the physician to the ethics committee D. Refuse to dictate
    the operative reporty.
  context: >-
    Inadvertent Cutting of Tendon is a complication, it should be in the
    Operative Reports The resident must put this complication in the operative
    report and disscuss it with the patient. If there was no harm to the patent
    and correction was done then theres nothing major for worry. But disclosing
    this as per ethical guidelines, is mandatory
  example_title: Example 1
---
# Model Card for MedGENIE-fid-flan-t5-base-medqa

MedGENIE comprises a collection of language models designed to utilize generated contexts, rather than retrieved ones, for addressing multiple-choice open-domain questions in the medical field. Specifically, **MedGENIE-fid-flan-t5-base-medqa** is a *fusion-in-decoder* (FID) model based on [flan-t5-base](https://huggingface.co/google/flan-t5-base), trained on the [MedQA-USMLE](https://huggingface.co/datasets/disi-unibo-nlp/medqa-5-opt-MedGENIE) dataset and grounded on artificial contexts generated by [PMC-LLaMA-13B](https://huggingface.co/axiong/PMC_LLaMA_13B). This model achieves a new *state-of-the-art* (SOTA) performance over the corresponding test set.

## Model description

- **Language(s) (NLP):** English
- **License:** MIT
- **Finetuned from model:** [google/flan-t5-base](https://huggingface.co/google/flan-t5-base)
- **Repository:** https://github.com/disi-unibo-nlp/medgenie
- **Paper:** [To Generate or to Retrieve? On the Effectiveness of Artificial Contexts for Medical Open-Domain Question Answering](https://arxiv.org/abs/2403.01924)

## Performance

At the time of release (February 2024), **MedGENIE-fid-flan-t5-base-medqa** is a new lightweight SOTA model on MedQA-USMLE benchmark:

| Model                            | Ground (Source)    | Learning                  | Params          | Accuracy (↓)             |
|----------------------------------|--------------------|---------------------------|-----------------|-------------------------------|
| **MedGENIE-FID-Flan-T5**         | **G (PMC-LLaMA)**      | **Fine-tuned**                | **250M**            | **53.1**                      |
| Codex <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>                   | &empty;            | 0-zhot                    | 175B            | 52.5                          |
| Codex <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>                  | R (Wikipedia)      | 0-shot                    | 175B            | 52.5                          |
| GPT-3.5-Turbo <small>([Yang et al.](https://arxiv.org/abs/2309.02233))</small>           | R (Wikipedia)      | k-shot                    | --              | 52.3                          |
| MEDITRON <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small>                | &empty;            | Fine-tuned                | 7B              | 52.0                          |
| BioMistral DARE <small> ([Labrak et al.](https://arxiv.org/abs/2402.10373)) </small>                     | &empty;            | Fine-tuned                    | 7B            | 51.1                          |
| BioMistral  <small> ([Labrak et al.](https://arxiv.org/abs/2402.10373)) </small>                      | &empty;            | Fine-tuned                    | 7B            | 50.6                         |
| Zephyr-&beta;                   | R (MedWiki)        | 2-shot                    | 7B              | 50.4                          |
| BioMedGPT <small>([Luo et al.](https://arxiv.org/abs/2308.09442v2))</small>              | &empty;            | k-shot                    | 10B             | 50.4                          |
| BioMedLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>               | &empty;            | Fine-tuned                | 2.7B            | 50.3                          |
| PMC-LLaMA <small>(awq 4 bit)</small>             | &empty;            | Fine-tuned                | 13B             | 50.2                          |
| LLaMA-2 <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small>              | &empty;            | Fine-tuned                | 7B              | 49.6                          |
| Zephyr-&beta;                  | &empty;            | 2-shot                    | 7B              | 49.6                          |
| Zephyr-&beta; <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small>          | &empty;            | 3-shot                    | 7B              | 49.2                          |
| PMC-LLaMA <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small>              | &empty;            | Fine-tuned                | 7B              | 49.2                          |
| DRAGON <small>([Yasunaga et al.](https://arxiv.org/abs/2210.09338))</small>                  | R (UMLS)           | Fine-tuned                | 360M            | 47.5                          |
| InstructGPT <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>             | R (Wikipedia)      | 0-shot                    | 175B            | 47.3                          |
| BioMistral DARE <small> ([Labrak et al.](https://arxiv.org/abs/2402.10373)) </small>                      | &empty;            | 3-shot                    | 7B            | 47.0                          |
| Flan-PaLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>             | &empty;            | 5-shot                    | 62B             | 46.1                          |
| InstructGPT <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>             | &empty;            | 0-shot                    | 175B            | 46.0                          |
| VOD <small>([Liévin et al. 2023](https://arxiv.org/abs/2210.06345))</small>                    | R (MedWiki)        | Fine-tuned                | 220M            | 45.8                          |
| Vicuna 1.3 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>              | &empty;            | 0-shot                    | 33B             | 45.2                          |
| BioLinkBERT <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>             | &empty;            | Fine-tuned                | 340M            | 45.1                          |
| Mistral-Instruct                 | R (MedWiki)        | 2-shot                    | 7B              | 45.1                          |
| BioMistral <small> ([Labrak et al.](https://arxiv.org/abs/2402.10373)) </small>                   | &empty;            | 3-shot                    | 7B            | 44.4                          |
| Galactica                        | &empty;            | 0-shot                    | 120B            | 44.4                          |
| LLaMA-2 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>                 | &empty;            | 0-shot                    | 70B             | 43.4                          |
| BioReader <small>([Frisoni et al.](https://aclanthology.org/2022.emnlp-main.390/))</small>               | R (PubMed-RCT)     | Fine-tuned                | 230M            | 43.0                          |
| Guanaco <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>             | &empty;            | 0-shot                    | 33B             | 42.9                          |
| LLaMA-2-chat <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>          | &empty;            | 0-shot                    | 70B             | 42.3                          |
| Vicuna 1.5 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>              | &empty;            | 0-shot                    | 65B             | 41.6                          |
| Mistral-Instruct <small>([Chen et al.](https://arxiv.org/abs/2311.16079))</small>        | &empty;            | 3-shot                    | 7B              | 41.1                          |
| PaLM <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>                 | &empty;            | 5-shot                    | 62B             | 40.9                          |
| Guanaco <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>             | &empty;            | 0-shot                    | 65B             | 40.8                          |
| Falcon-Instruct <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>         | &empty;            | 0-shot                    | 40B             | 39.0                          |
| Vicuna 1.3 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>              | &empty;            | 0-shot                    | 13B             | 38.7                          |
| GreaseLM <small>([Zhang et al.](https://arxiv.org/abs/2201.08860))</small>              | R (UMLS)           | Fine-tuned                | 359M            | 38.5                          |
| PubMedBERT <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>              | &empty;            | Fine-tuned                | 110M            | 38.1                          |
| QA-GNN <small>([Yasunaga et al.](https://arxiv.org/abs/2104.06378))</small>               | R (UMLS)           | Fine-tuned                | 360M            | 38.0                          |
| LLaMA-2 <small>([Yang et al.](https://arxiv.org/abs/2309.02233))</small>               | R (Wikipedia)      | k-shot                    | 13B             | 37.6                          |
| LLaMA-2-chat                     | R (MedWiki)        | 2-shot                    | 7B              | 37.2                          |
| LLaMA-2-chat                     | &empty;            | 2-shot                    | 7B              | 37.2                          |
| BioBERT <small>([Lee et al.](https://arxiv.org/abs/1901.08746))</small>                 | &empty;            | Fine-tuned                | 110M            | 36.7                          |
| MTP-Instruct <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>          | &empty;            | 0-shot                    | 30B             | 35.1                          |
| GPT-Neo <small>([Singhal et al.](https://arxiv.org/abs/2212.13138))</small>                 | &empty;            | Fine-tuned                | 2.5B            | 33.3                          |
| LLaMa-2-chat <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>         | &empty;            | 0-shot                    | 13B             | 32.2                          |
| LLaMa-2 <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>                 | &empty;            | 0-shot                    | 13B             | 31.1                          |
| GPT-NeoX <small>([Liévin et al.](https://arxiv.org/abs/2207.08143))</small>               | &empty;            | 0-shot                    | 20B             | 26.9                          |    

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- n_context: 5
- per_gpu_batch_size: 1
- accumulation_steps: 4 
- total_steps: 40,712
- eval_freq: 10,178
- optimizer: AdamW 
- scheduler: linear
- weight_decay: 0.01
- warmup_ratio: 0.1
- text_maxlength: 1024

### Bias, Risk and Limitation

Our model is trained on artificially generated contextual documents, which might inadvertently magnify inherent biases and depart from clinical and societal norms. This could lead to the spread of convincing medical misinformation. To mitigate this risk, we recommend a cautious approach: domain experts should manually review any output before real-world use. This ethical safeguard is crucial to prevent the dissemination of potentially erroneous or misleading information, particularly within clinical and scientific circles.

## Citation

If you find MedGENIE-fid-flan-t5-base-medqa is useful in your work, please cite it with:

```
@misc{frisoni2024generate,
      title={To Generate or to Retrieve? On the Effectiveness of Artificial Contexts for Medical Open-Domain Question Answering}, 
      author={Giacomo Frisoni and Alessio Cocchieri and Alex Presepi and Gianluca Moro and Zaiqiao Meng},
      year={2024},
      eprint={2403.01924},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```