Bart Cant commited on
Commit ·
6671112
1
Parent(s): 9168834
initial comit
Browse files- app.py +146 -0
- requirements.txt +7 -0
app.py
ADDED
|
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import gradio as gr
|
| 3 |
+
|
| 4 |
+
# Get Hugging Face API Key
|
| 5 |
+
huggingface_token = os.environ.get('HUGGINGFACE_API_KEY')
|
| 6 |
+
|
| 7 |
+
# PubMed Fetcher
|
| 8 |
+
from pymed import PubMed
|
| 9 |
+
from typing import List
|
| 10 |
+
from haystack import component
|
| 11 |
+
from haystack import Document
|
| 12 |
+
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
def predict( message, history):
|
| 16 |
+
|
| 17 |
+
pubmed = PubMed(tool="Tydei", email="info@tydei.io")
|
| 18 |
+
|
| 19 |
+
def documentize(article):
|
| 20 |
+
return Document(content=article.abstract, meta={'title': article.title, 'keywords': article.keywords, 'journal': article.journal})
|
| 21 |
+
|
| 22 |
+
@component
|
| 23 |
+
class PubMedFetcher():
|
| 24 |
+
@component.output_types(articles=List[Document])
|
| 25 |
+
def run(self, queries: list[str]):
|
| 26 |
+
cleaned_queries = queries[0].strip().split('\n')
|
| 27 |
+
articles = []
|
| 28 |
+
try:
|
| 29 |
+
for query in cleaned_queries:
|
| 30 |
+
response = pubmed.query(query, max_results=3)
|
| 31 |
+
documents = [documentize(article) for article in response]
|
| 32 |
+
articles.extend(documents)
|
| 33 |
+
except Exception as e:
|
| 34 |
+
print(e)
|
| 35 |
+
print(f"Couldn't fetch articles for queries: {queries}")
|
| 36 |
+
results = {'articles': articles}
|
| 37 |
+
print("-----------------------------")
|
| 38 |
+
print ("results : ", results)
|
| 39 |
+
# Initialize an empty list to store all keywords
|
| 40 |
+
all_keywords = []
|
| 41 |
+
|
| 42 |
+
# Iterate over each Document in the articles list
|
| 43 |
+
for document in articles:
|
| 44 |
+
# Extract the keywords from the meta attribute of each Document
|
| 45 |
+
keywords = document.meta.get('keywords', [])
|
| 46 |
+
# Extend the all_keywords list with the extracted keywords
|
| 47 |
+
all_keywords.extend(keywords)
|
| 48 |
+
|
| 49 |
+
# Add the extracted keywords to the results dictionary
|
| 50 |
+
results['keywords'] = all_keywords
|
| 51 |
+
# Print all extracted keywords
|
| 52 |
+
print("Extracted Keywords:", all_keywords)
|
| 53 |
+
|
| 54 |
+
return results
|
| 55 |
+
|
| 56 |
+
# Pipeline Setup
|
| 57 |
+
from haystack.components.generators import HuggingFaceTGIGenerator
|
| 58 |
+
from haystack import Pipeline
|
| 59 |
+
from haystack.components.builders.prompt_builder import PromptBuilder
|
| 60 |
+
|
| 61 |
+
keyword_llm = HuggingFaceTGIGenerator("mistralai/Mixtral-8x7B-Instruct-v0.1", token=huggingface_token)
|
| 62 |
+
keyword_llm.warm_up()
|
| 63 |
+
|
| 64 |
+
llm = HuggingFaceTGIGenerator("mistralai/Mixtral-8x7B-Instruct-v0.1", token=huggingface_token)
|
| 65 |
+
llm.warm_up()
|
| 66 |
+
|
| 67 |
+
from haystack import Pipeline
|
| 68 |
+
from haystack.components.builders.prompt_builder import PromptBuilder
|
| 69 |
+
|
| 70 |
+
keyword_prompt_template = """
|
| 71 |
+
Your task is to convert the follwing question into 5 keywords that can be used to find relevant medical research papers on PubMed.
|
| 72 |
+
|
| 73 |
+
---
|
| 74 |
+
question: {{ question }}
|
| 75 |
+
keywords:
|
| 76 |
+
"""
|
| 77 |
+
|
| 78 |
+
prompt_template = """
|
| 79 |
+
Answer the question truthfully based on the given documents.
|
| 80 |
+
Only use data from the given documents to answer the question.
|
| 81 |
+
|
| 82 |
+
q: {{ question }}
|
| 83 |
+
|
| 84 |
+
Articles:
|
| 85 |
+
{% for article in articles %}
|
| 86 |
+
{{article.content}}
|
| 87 |
+
keywords: {{article.meta['keywords']}}
|
| 88 |
+
title: {{article.meta['title']}}
|
| 89 |
+
journal: {{article.journal}}
|
| 90 |
+
{% endfor %}
|
| 91 |
+
|
| 92 |
+
Only provide an answer based on the information give. Do not make anything up.
|
| 93 |
+
|
| 94 |
+
At the end of your answer create a Source List with the
|
| 95 |
+
name of Article title and Journal. Retrieve your information from the article meta data
|
| 96 |
+
|
| 97 |
+
DO not repeat yourself in the answer and ONLY use information from the given articles. DO NOT USE Your Own Knowledge base or you will be punished
|
| 98 |
+
"""
|
| 99 |
+
|
| 100 |
+
keyword_prompt_builder = PromptBuilder(template=keyword_prompt_template)
|
| 101 |
+
prompt_builder = PromptBuilder(template=prompt_template)
|
| 102 |
+
fetcher = PubMedFetcher()
|
| 103 |
+
|
| 104 |
+
pipe = Pipeline()
|
| 105 |
+
pipe.add_component("keyword_prompt_builder", keyword_prompt_builder)
|
| 106 |
+
pipe.add_component("keyword_llm", keyword_llm)
|
| 107 |
+
pipe.add_component("pubmed_fetcher", fetcher)
|
| 108 |
+
pipe.add_component("prompt_builder", prompt_builder)
|
| 109 |
+
pipe.add_component("llm", llm)
|
| 110 |
+
|
| 111 |
+
pipe.connect("keyword_prompt_builder.prompt", "keyword_llm.prompt")
|
| 112 |
+
pipe.connect("keyword_llm.replies", "pubmed_fetcher.queries")
|
| 113 |
+
pipe.connect("pubmed_fetcher.articles", "prompt_builder.articles")
|
| 114 |
+
pipe.connect("prompt_builder.prompt", "llm.prompt")
|
| 115 |
+
|
| 116 |
+
|
| 117 |
+
# Function to Ask Questions
|
| 118 |
+
def ask(question):
|
| 119 |
+
result = pipe.run(data={"keyword_prompt_builder":{"question":question},
|
| 120 |
+
"prompt_builder":{"question": question},
|
| 121 |
+
"llm":{"generation_kwargs": {"max_new_tokens": 1500}}})
|
| 122 |
+
|
| 123 |
+
print("-----------------------------")
|
| 124 |
+
print("Question is: ", question)
|
| 125 |
+
print("-----------------------------")
|
| 126 |
+
print(result)
|
| 127 |
+
output = result['llm']['replies'][0]
|
| 128 |
+
keywords = result['pubmed_fetcher']['keywords']
|
| 129 |
+
return keywords, output
|
| 130 |
+
|
| 131 |
+
|
| 132 |
+
keywords , output= ask(message)
|
| 133 |
+
|
| 134 |
+
return keywords, output
|
| 135 |
+
|
| 136 |
+
# Launch Gradio Interface
|
| 137 |
+
|
| 138 |
+
interface = gr.Interface(
|
| 139 |
+
fn=predict,
|
| 140 |
+
inputs= [gr.Textbox(label="Research Pubmed...", lines=2, placeholder="Type Question here..."),
|
| 141 |
+
gr.Dropdown(["ortho", "anthroscopy", "osteotomu", "fracture repair"], value=["spinal fusion", "bone grafting"], multiselect=True, label="Keyword Entry")],
|
| 142 |
+
outputs=[gr.Textbox(label="Keywords used in researching articles", lines=5),
|
| 143 |
+
gr.Textbox(label="Pubmed Insights", lines=20)]
|
| 144 |
+
)
|
| 145 |
+
|
| 146 |
+
interface.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
pymongo==4.6.1
|
| 2 |
+
langchain-core==0.1.9
|
| 3 |
+
langchain==0.1.0
|
| 4 |
+
langchain-community==0.0.11
|
| 5 |
+
langchain-openai==0.0.2
|
| 6 |
+
openai==1.6.1
|
| 7 |
+
tiktoken==0.5.2
|