Spaces:
Sleeping
Sleeping
Updated to incl SERPER_API and Openai
Browse files
app.py
CHANGED
@@ -1,25 +1,20 @@
|
|
1 |
-
|
|
|
2 |
from transformers import BertTokenizer, BertForSequenceClassification, TrainingArguments, Trainer
|
3 |
import openai
|
|
|
4 |
import faiss
|
5 |
import numpy as np
|
|
|
6 |
|
7 |
-
#
|
8 |
-
openai.api_key = "
|
|
|
9 |
|
10 |
-
# Load PubMedBERT tokenizer and model
|
11 |
tokenizer = BertTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
|
12 |
model = BertForSequenceClassification.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract", num_labels=2)
|
13 |
|
14 |
-
# Load the FDA dataset from Hugging Face
|
15 |
-
dataset = load_dataset("pretzinger/cdx-cleared-approved", split="train")
|
16 |
-
|
17 |
-
# Tokenize the dataset
|
18 |
-
def tokenize_function(example):
|
19 |
-
return tokenizer(example["text"], padding="max_length", truncation=True)
|
20 |
-
|
21 |
-
tokenized_dataset = dataset.map(tokenize_function, batched=True)
|
22 |
-
|
23 |
# FAISS setup for vector search (embedding-based memory)
|
24 |
dimension = 768 # PubMedBERT embedding size
|
25 |
index = faiss.IndexFlatL2(dimension)
|
@@ -40,43 +35,96 @@ def search_memory(query):
|
|
40 |
D, I = index.search(query_embedding, k=1) # Retrieve most similar past conversation
|
41 |
return I
|
42 |
|
43 |
-
# Function to handle FDA-
|
44 |
def handle_fda_query(query):
|
45 |
-
# If query requires specific FDA info, process it with PubMedBERT
|
46 |
inputs = tokenizer(query, return_tensors="pt", padding="max_length", truncation=True)
|
47 |
outputs = model(**inputs)
|
48 |
logits = outputs.logits
|
49 |
-
# Process logits for classification or output a meaningful response
|
50 |
response = "Processed FDA-related query via PubMedBERT"
|
51 |
return response
|
52 |
|
53 |
-
# Function to handle general queries using GPT-
|
54 |
def handle_openai_query(prompt):
|
55 |
response = openai.Completion.create(
|
56 |
-
engine="gpt-
|
57 |
prompt=prompt,
|
58 |
max_tokens=100
|
59 |
)
|
60 |
return response.choices[0].text.strip()
|
61 |
|
62 |
-
#
|
63 |
-
def
|
64 |
-
|
65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
|
67 |
if "FDA" in openai_response or "regulatory" in openai_response:
|
68 |
# Search past conversations/memory using FAISS
|
69 |
-
memory_index = search_memory(
|
70 |
if memory_index:
|
71 |
return f"Found relevant past memory: {past_conversation}" # Return past context from memory
|
72 |
|
73 |
# If no memory match, proceed with PubMedBERT
|
74 |
-
return handle_fda_query(
|
75 |
-
|
76 |
-
#
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
response = assistant(query)
|
82 |
-
print(response)
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from huggingface_hub import InferenceClient
|
3 |
from transformers import BertTokenizer, BertForSequenceClassification, TrainingArguments, Trainer
|
4 |
import openai
|
5 |
+
import os
|
6 |
import faiss
|
7 |
import numpy as np
|
8 |
+
import requests
|
9 |
|
10 |
+
# Load OpenAI and Serper API keys from Hugging Face secrets
|
11 |
+
openai.api_key = os.getenv("OPENAI_API_KEY") # Ensure the OpenAI API key is pulled correctly
|
12 |
+
serper_api_key = os.getenv("SERPER_API_KEY") # Ensure the Serper API key is pulled correctly
|
13 |
|
14 |
+
# Load PubMedBERT tokenizer and model for FDA-related processing
|
15 |
tokenizer = BertTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
|
16 |
model = BertForSequenceClassification.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract", num_labels=2)
|
17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
# FAISS setup for vector search (embedding-based memory)
|
19 |
dimension = 768 # PubMedBERT embedding size
|
20 |
index = faiss.IndexFlatL2(dimension)
|
|
|
35 |
D, I = index.search(query_embedding, k=1) # Retrieve most similar past conversation
|
36 |
return I
|
37 |
|
38 |
+
# Function to handle FDA-specific queries with PubMedBERT
|
39 |
def handle_fda_query(query):
|
|
|
40 |
inputs = tokenizer(query, return_tensors="pt", padding="max_length", truncation=True)
|
41 |
outputs = model(**inputs)
|
42 |
logits = outputs.logits
|
|
|
43 |
response = "Processed FDA-related query via PubMedBERT"
|
44 |
return response
|
45 |
|
46 |
+
# Function to handle general queries using GPT-4o
|
47 |
def handle_openai_query(prompt):
|
48 |
response = openai.Completion.create(
|
49 |
+
engine="gpt-4o", # Using GPT-4o as per instruction
|
50 |
prompt=prompt,
|
51 |
max_tokens=100
|
52 |
)
|
53 |
return response.choices[0].text.strip()
|
54 |
|
55 |
+
# Web search with Serper API
|
56 |
+
def web_search(query):
|
57 |
+
url = f"https://google.serper.dev/search"
|
58 |
+
headers = {
|
59 |
+
"X-API-KEY": serper_api_key
|
60 |
+
}
|
61 |
+
params = {
|
62 |
+
"q": query
|
63 |
+
}
|
64 |
+
response = requests.get(url, headers=headers, params=params)
|
65 |
+
return response.json()
|
66 |
+
|
67 |
+
# Main assistant function that delegates to either OpenAI, PubMedBERT, or Serper (web search)
|
68 |
+
def respond(
|
69 |
+
message,
|
70 |
+
history: list[tuple[str, str]],
|
71 |
+
system_message,
|
72 |
+
max_tokens,
|
73 |
+
temperature,
|
74 |
+
top_p,
|
75 |
+
):
|
76 |
+
# Prepare the context for OpenAI and PubMedBERT
|
77 |
+
messages = [{"role": "system", "content": system_message}]
|
78 |
+
|
79 |
+
for val in history:
|
80 |
+
if val[0]:
|
81 |
+
messages.append({"role": "user", "content": val[0]})
|
82 |
+
if val[1]:
|
83 |
+
messages.append({"role": "assistant", "content": val[1]})
|
84 |
+
|
85 |
+
messages.append({"role": "user", "content": message})
|
86 |
+
|
87 |
+
# Check if the query is related to FDA
|
88 |
+
openai_response = handle_openai_query(f"Is this query FDA-related: {message}")
|
89 |
|
90 |
if "FDA" in openai_response or "regulatory" in openai_response:
|
91 |
# Search past conversations/memory using FAISS
|
92 |
+
memory_index = search_memory(message)
|
93 |
if memory_index:
|
94 |
return f"Found relevant past memory: {past_conversation}" # Return past context from memory
|
95 |
|
96 |
# If no memory match, proceed with PubMedBERT
|
97 |
+
return handle_fda_query(message)
|
98 |
+
|
99 |
+
# If query asks for a web search, perform web search
|
100 |
+
if "search the web" in message.lower():
|
101 |
+
return web_search(message)
|
102 |
+
|
103 |
+
# General conversational handling with GPT-4o
|
104 |
+
response = ""
|
105 |
+
for message in client.chat_completion(
|
106 |
+
messages,
|
107 |
+
max_tokens=max_tokens,
|
108 |
+
stream=True,
|
109 |
+
temperature=temperature,
|
110 |
+
top_p=top_p,
|
111 |
+
):
|
112 |
+
token = message.choices[0].delta.content
|
113 |
+
|
114 |
+
response += token
|
115 |
+
yield response
|
116 |
+
|
117 |
+
|
118 |
+
# Create Gradio ChatInterface for interaction
|
119 |
+
demo = gr.ChatInterface(
|
120 |
+
respond,
|
121 |
+
additional_inputs=[
|
122 |
+
gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
|
123 |
+
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
|
124 |
+
gr.Slider(minimum=0.1, maximum 4.0, value=0.7, step=0.1, label="Temperature"),
|
125 |
+
gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
|
126 |
+
],
|
127 |
+
)
|
128 |
|
129 |
+
if __name__ == "__main__":
|
130 |
+
demo.launch()
|
|
|
|