kbressem commited on
Commit
6862c63
1 Parent(s): 7e967e4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +72 -0
README.md CHANGED
@@ -1,3 +1,75 @@
1
  ---
2
  license: cc
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc
3
+ language:
4
+ - en
5
+ library_name: transformers
6
+ pipeline_tag: text-generation
7
+ tags:
8
+ - medical
9
  ---
10
+ # MedAlpaca 13b
11
+
12
+
13
+ ## Table of Contents
14
+
15
+ [Model Description](#model-description)
16
+ - [Architecture](#architecture)
17
+ - [Training Data](#trainig-data)
18
+ [Model Usage](#model-usage)
19
+ [Limitations](#limitations)
20
+
21
+ ## Model Description
22
+ ### Architecture
23
+ `medalpaca-13b` is a large language model specifically fine-tuned for medical domain tasks.
24
+ It is based on LLaMA (Large Language Model Meta AI) and contains 13 billion parameters.
25
+ The primary goal of this model is to improve question-answering and medical dialogue tasks.
26
+
27
+ ### Training Data
28
+ The training data for this project was sourced from various resources.
29
+ Firstly, we used Anki flashcards to automatically generate questions,
30
+ from the front of the cards and anwers from the back of the card.
31
+ Secondly, we generated medical question-answer pairs from [Wikidoc](https://www.wikidoc.org/index.php/Main_Page).
32
+ We extracted paragraphs with relevant headings, and used Chat-GPT 3.5
33
+ to generate questions from the headings and using the corresponding paragraphs
34
+ as answers. This dataset is still under development and we believe
35
+ that approximately 70% of these question answer pairs are factual correct.
36
+ Thirdly, we used StackExchange to extract question-answer pairs, taking the
37
+ top-rated question from five categories: Academia, Bioinformatics, Biology,
38
+ Fitness, and Health. Additionally, we used a dataset from [ChatDoctor](https://arxiv.org/abs/2303.14070)
39
+ consisting of 200,000 question-answer pairs, available at https://github.com/Kent0n-Li/ChatDoctor.
40
+
41
+ | Source | n items |
42
+ |------------------------------|--------|
43
+ | ChatDoc large | 200000 |
44
+ | wikidoc | 67704 |
45
+ | Stackexchange academia | 40865 |
46
+ | Anki flashcards | 33955 |
47
+ | Stackexchange biology | 27887 |
48
+ | Stackexchange fitness | 9833 |
49
+ | Stackexchange health | 7721 |
50
+ | Wikidoc patient information | 5942 |
51
+ | Stackexchange bioinformatics | 5407 |
52
+
53
+ ## Model Usage
54
+ To evaluate the performance of the model on a specific dataset, you can use the Hugging Face Transformers library's built-in evaluation scripts. Please refer to the evaluation guide for more information.
55
+ Inference
56
+
57
+ You can use the model for inference tasks like question-answering and medical dialogues using the Hugging Face Transformers library. Here's an example of how to use the model for a question-answering task:
58
+
59
+ ```python
60
+
61
+ from transformers import pipeline
62
+
63
+ qa_pipeline = pipeline("question-answering", model="medalpaca/medalpaca-7b", tokenizer="medalpaca/medalpaca-7b")
64
+ question = "What are the symptoms of diabetes?"
65
+ context = "Diabetes is a metabolic disease that causes high blood sugar. The symptoms include increased thirst, frequent urination, and unexplained weight loss."
66
+ answer = qa_pipeline({"question": question, "context": context})
67
+ print(answer)
68
+ ```
69
+
70
+ ## Limitations
71
+ The model may not perform effectively outside the scope of the medical domain.
72
+ The training data primarily targets the knowledge level of medical students,
73
+ which may result in limitations when addressing the needs of board-certified physicians.
74
+ The model has not been tested in real-world applications, so its efficacy and accuracy are currently unknown.
75
+ It should never be used as a substitute for a doctor's opinion and must be treated as a research tool only.