Datasets:
Update README.md
Browse files
README.md
CHANGED
@@ -4,52 +4,59 @@ configs:
|
|
4 |
- config_name: english-full
|
5 |
data_files:
|
6 |
- split: train
|
7 |
-
path:
|
8 |
- split: test
|
9 |
-
path:
|
10 |
- config_name: english-qi
|
11 |
data_files:
|
12 |
- split: train
|
13 |
-
path:
|
14 |
- split: test
|
15 |
-
path:
|
16 |
- config_name: english-dp
|
17 |
data_files:
|
18 |
- split: train
|
19 |
-
path:
|
20 |
- split: test
|
21 |
-
path:
|
22 |
- config_name: english-custom
|
23 |
data_files:
|
24 |
- split: train
|
25 |
-
path:
|
26 |
- split: test
|
27 |
-
path:
|
28 |
- config_name: french-full
|
29 |
data_files:
|
30 |
- split: train
|
31 |
-
path:
|
32 |
- split: test
|
33 |
-
path:
|
34 |
- config_name: french-qi
|
35 |
data_files:
|
36 |
- split: train
|
37 |
-
path:
|
38 |
- split: test
|
39 |
-
path:
|
40 |
- config_name: french-dp
|
41 |
data_files:
|
42 |
- split: train
|
43 |
-
path:
|
44 |
- split: test
|
45 |
-
path:
|
46 |
- config_name: french-custom
|
47 |
data_files:
|
48 |
- split: train
|
49 |
-
path:
|
50 |
- split: test
|
51 |
-
path:
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
---
|
54 |
|
55 |
|
@@ -61,12 +68,6 @@ Paper: [https://arxiv.org/abs/2405.14654](https://arxiv.org/abs/2405.14654)
|
|
61 |
|
62 |
## Dataset Details
|
63 |
|
64 |
-
In the expanding field of language model applications, medical knowledge representation remains a significant challenge due to the specialized nature of the domain.
|
65 |
-
Large language models, such as GPT-4, obtain reasonable scores on medical question-answering tasks, but smaller models are far behind.
|
66 |
-
In this work, we introduce a method to improve the proficiency of a small language model in the medical domain by employing a two-fold approach.
|
67 |
-
We first fine-tune the model on a corpus of medical textbooks. Then, we use GPT-4 to generate questions similar to the downstream task, prompted with textbook knowledge, and use them to fine-tune the model.
|
68 |
-
We show the benefits of our training strategy on a medical answering question dataset.
|
69 |
-
The study's findings highlight the potential of small language models in the medical domain when appropriately fine-tuned.
|
70 |
|
71 |
### Dataset Description
|
72 |
|
@@ -78,7 +79,7 @@ The questions can be:
|
|
78 |
|
79 |
There are two versions of this dataset: the **french** and the **english** versions.
|
80 |
The **french** split is the original dataset version.
|
81 |
-
|
82 |
|
83 |
### Use this dataset
|
84 |
|
@@ -107,7 +108,14 @@ ds_english_custom = load_dataset("raidium/ECN-QA", "english-custom")
|
|
107 |
- **Developed by:** Raidium
|
108 |
- **License:** Apache 2.0
|
109 |
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
|
112 |
<!-- Provide the basic links for the model. -->
|
113 |
|
|
|
4 |
- config_name: english-full
|
5 |
data_files:
|
6 |
- split: train
|
7 |
+
path: english/train.jsonl
|
8 |
- split: test
|
9 |
+
path: english/test.jsonl
|
10 |
- config_name: english-qi
|
11 |
data_files:
|
12 |
- split: train
|
13 |
+
path: english/qi_train.jsonl
|
14 |
- split: test
|
15 |
+
path: english/qi_test.jsonl
|
16 |
- config_name: english-dp
|
17 |
data_files:
|
18 |
- split: train
|
19 |
+
path: english/dp_train.jsonl
|
20 |
- split: test
|
21 |
+
path: english/dp_test.jsonl
|
22 |
- config_name: english-custom
|
23 |
data_files:
|
24 |
- split: train
|
25 |
+
path: english/custom_train.jsonl
|
26 |
- split: test
|
27 |
+
path: english/custom_test.jsonl
|
28 |
- config_name: french-full
|
29 |
data_files:
|
30 |
- split: train
|
31 |
+
path: french/train.jsonl
|
32 |
- split: test
|
33 |
+
path: french/test.jsonl
|
34 |
- config_name: french-qi
|
35 |
data_files:
|
36 |
- split: train
|
37 |
+
path: french/qi_train.jsonl
|
38 |
- split: test
|
39 |
+
path: french/qi_test.jsonl
|
40 |
- config_name: french-dp
|
41 |
data_files:
|
42 |
- split: train
|
43 |
+
path: french/dp_train.jsonl
|
44 |
- split: test
|
45 |
+
path: french/dp_test.jsonl
|
46 |
- config_name: french-custom
|
47 |
data_files:
|
48 |
- split: train
|
49 |
+
path: french/custom_train.jsonl
|
50 |
- split: test
|
51 |
+
path: french/custom_test.jsonl
|
52 |
+
task_categories:
|
53 |
+
- multiple-choice
|
54 |
+
language:
|
55 |
+
- fr
|
56 |
+
- en
|
57 |
+
tags:
|
58 |
+
- medical question answering
|
59 |
+
pretty_name: ecn-qa
|
60 |
---
|
61 |
|
62 |
|
|
|
68 |
|
69 |
## Dataset Details
|
70 |
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
|
72 |
### Dataset Description
|
73 |
|
|
|
79 |
|
80 |
There are two versions of this dataset: the **french** and the **english** versions.
|
81 |
The **french** split is the original dataset version.
|
82 |
+
|
83 |
|
84 |
### Use this dataset
|
85 |
|
|
|
108 |
- **Developed by:** Raidium
|
109 |
- **License:** Apache 2.0
|
110 |
|
111 |
+
|
112 |
+
### Warnings
|
113 |
+
|
114 |
+
- Some questions require images to be answered. They have not been filtered out in this dataset, to it is impossible to get 100% accuracy on this dataset.
|
115 |
+
- The english version is an automated translation (using Azure Translation api), hence, it might contain translation errors.
|
116 |
+
|
117 |
+
|
118 |
+
### Dataset Sources
|
119 |
|
120 |
<!-- Provide the basic links for the model. -->
|
121 |
|