Spaces:
Sleeping
Sleeping
jocko
commited on
Commit
Β·
0a89a37
1
Parent(s):
4a259f2
copy updates of mult modal
Browse files- README.md +3 -4
- src/streamlit_app.py +103 -140
README.md
CHANGED
@@ -1,5 +1,5 @@
|
|
1 |
---
|
2 |
-
title: Dr Q
|
3 |
emoji: π
|
4 |
colorFrom: red
|
5 |
colorTo: red
|
@@ -8,12 +8,11 @@ app_port: 8501
|
|
8 |
tags:
|
9 |
- streamlit
|
10 |
pinned: false
|
11 |
-
short_description:
|
12 |
---
|
13 |
-
|
14 |
# Welcome to Streamlit!
|
15 |
|
16 |
Edit `/src/streamlit_app.py` to customize this app to your heart's desire. :heart:
|
17 |
|
18 |
If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
|
19 |
-
forums](https://discuss.streamlit.io).
|
|
|
1 |
---
|
2 |
+
title: Dr Q
|
3 |
emoji: π
|
4 |
colorFrom: red
|
5 |
colorTo: red
|
|
|
8 |
tags:
|
9 |
- streamlit
|
10 |
pinned: false
|
11 |
+
short_description: Multimodal medical chatbot
|
12 |
---
|
|
|
13 |
# Welcome to Streamlit!
|
14 |
|
15 |
Edit `/src/streamlit_app.py` to customize this app to your heart's desire. :heart:
|
16 |
|
17 |
If you have any questions, checkout our [documentation](https://docs.streamlit.io) and [community
|
18 |
+
forums](https://discuss.streamlit.io).
|
src/streamlit_app.py
CHANGED
@@ -1,164 +1,127 @@
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
|
3 |
-
#
|
4 |
-
|
5 |
-
os.environ["
|
6 |
-
os.environ["
|
7 |
-
os.environ["
|
8 |
-
os.environ["
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
import
|
20 |
import torch
|
21 |
-
import openai
|
22 |
-
import os
|
23 |
from sentence_transformers import SentenceTransformer, util
|
24 |
-
import
|
25 |
-
from
|
26 |
-
|
27 |
-
|
28 |
-
# Set the API key
|
29 |
-
client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
30 |
-
# openai.api_key = os.getenv("OPENAI_API_KEY")
|
31 |
-
# REMEDI_PATH = "ReMeDi-base.json"
|
32 |
-
BASE_DIR = Path(__file__).parent
|
33 |
-
REMEDI_PATH = BASE_DIR / "ReMeDi-base.json"
|
34 |
-
|
35 |
-
# Check if file exists
|
36 |
-
if not REMEDI_PATH.exists():
|
37 |
-
raise FileNotFoundError(f"β File not found: {REMEDI_PATH}")
|
38 |
-
|
39 |
-
# Load the file
|
40 |
-
with open(REMEDI_PATH, "r", encoding="utf-8") as f:
|
41 |
-
data = json.load(f)
|
42 |
-
|
43 |
-
|
44 |
-
# === LOAD MODEL ===
|
45 |
-
@st.cache_resource
|
46 |
-
def load_model():
|
47 |
-
return SentenceTransformer("all-MiniLM-L6-v2")
|
48 |
-
# return model
|
49 |
-
|
50 |
-
|
51 |
-
@st.cache_resource
|
52 |
-
def load_data():
|
53 |
-
with open(REMEDI_PATH, "r", encoding="utf-8") as f:
|
54 |
-
data = json.load(f)
|
55 |
-
dialogue_pairs = []
|
56 |
-
for conversation in data:
|
57 |
-
turns = conversation["information"]
|
58 |
-
for i in range(len(turns) - 1):
|
59 |
-
if turns[i]["role"] == "patient" and turns[i + 1]["role"] == "doctor":
|
60 |
-
dialogue_pairs.append({
|
61 |
-
"patient": turns[i]["sentence"],
|
62 |
-
"doctor": turns[i + 1]["sentence"]
|
63 |
-
})
|
64 |
-
return dialogue_pairs
|
65 |
-
|
66 |
-
|
67 |
-
@st.cache_data
|
68 |
-
def build_embeddings(dialogue_pairs, _model):
|
69 |
-
patient_sentences = [pair["patient"] for pair in dialogue_pairs]
|
70 |
-
embeddings = _model.encode(patient_sentences, convert_to_tensor=True)
|
71 |
-
return embeddings
|
72 |
-
|
73 |
-
|
74 |
-
# === TRANSLATE USING GPT ===
|
75 |
-
def translate_to_english(chinese_text):
|
76 |
-
prompt = f"Translate the following Chinese medical response to English:\n\n{chinese_text}"
|
77 |
-
try:
|
78 |
-
response = client.chat.completions.create(
|
79 |
-
model="gpt-4",
|
80 |
-
messages=[{"role": "user", "content": prompt}],
|
81 |
-
temperature=0.2
|
82 |
-
)
|
83 |
-
return response.choices[0].message.content
|
84 |
|
85 |
-
|
86 |
-
|
87 |
-
return f"Translation failed: {str(e)}"
|
88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
model="gpt-4", # or "gpt-3.5-turbo" to save credits
|
95 |
-
messages=[{"role": "user", "content": prompt}],
|
96 |
-
temperature=0.5
|
97 |
-
)
|
98 |
-
return response.choices[0].message.content
|
99 |
-
except Exception as e:
|
100 |
-
return f"GPT response failed: {str(e)}"
|
101 |
|
102 |
|
103 |
-
# === CHATBOT FUNCTION ===
|
104 |
-
def chatbot_response(user_input, _model, dialogue_pairs, patient_embeddings, top_k=1):
|
105 |
-
user_embedding = _model.encode(user_input, convert_to_tensor=True)
|
106 |
-
similarities = util.cos_sim(user_embedding, patient_embeddings)[0]
|
107 |
-
top_score, top_idx = torch.topk(similarities, k=1)
|
108 |
-
top_score = top_score.item()
|
109 |
-
top_idx = torch.topk(similarities, k=top_k).indices[0].item()
|
110 |
|
111 |
-
|
112 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
|
114 |
-
|
115 |
-
"matched_question": match["patient"],
|
116 |
-
"original_response": match["doctor"],
|
117 |
-
"translated_response": translated
|
118 |
-
# "similarity_score": top_score
|
119 |
-
}
|
120 |
|
|
|
|
|
121 |
|
122 |
-
#
|
123 |
-
|
124 |
-
st.title("π©Ί Dr_Q_bot - Medical Chatbot")
|
125 |
-
st.write("Ask about a symptom and get an example doctor response (translated from Chinese).")
|
126 |
|
127 |
-
#
|
128 |
-
|
129 |
-
dialogue_pairs = load_data()
|
130 |
-
patient_embeddings = build_embeddings(dialogue_pairs, model)
|
131 |
|
132 |
-
#
|
133 |
-
|
|
|
|
|
134 |
|
135 |
-
|
136 |
-
|
137 |
-
result = chatbot_response(user_input, model, dialogue_pairs, patient_embeddings)
|
138 |
-
gpt_response = gpt_direct_response(user_input)
|
139 |
|
140 |
-
|
141 |
-
|
142 |
|
143 |
-
|
144 |
|
145 |
-
|
146 |
-
|
147 |
-
|
|
|
148 |
|
149 |
-
|
150 |
-
|
|
|
|
|
|
|
151 |
|
152 |
-
|
153 |
-
st.
|
154 |
-
# else:
|
155 |
-
# st.warning("No close match found in dataset. Using GPT response only.")
|
156 |
|
157 |
-
#
|
158 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
|
160 |
-
|
|
|
|
|
161 |
|
162 |
-
|
163 |
-
st.warning(
|
164 |
-
"This chatbot uses real dialogue data for research and educational use only. Not a substitute for professional medical advice.")
|
|
|
1 |
+
# ================================
|
2 |
+
# β
Cache-Safe Multimodal App
|
3 |
+
# ================================
|
4 |
+
|
5 |
import os
|
6 |
|
7 |
+
# ====== Force all cache dirs to /tmp (writable in most environments) ======
|
8 |
+
CACHE_BASE = "/tmp/cache"
|
9 |
+
os.environ["HF_HOME"] = f"{CACHE_BASE}/hf_home"
|
10 |
+
os.environ["TRANSFORMERS_CACHE"] = f"{CACHE_BASE}/transformers"
|
11 |
+
os.environ["SENTENCE_TRANSFORMERS_HOME"] = f"{CACHE_BASE}/sentence_transformers"
|
12 |
+
os.environ["HF_DATASETS_CACHE"] = f"{CACHE_BASE}/hf_datasets"
|
13 |
+
os.environ["TORCH_HOME"] = f"{CACHE_BASE}/torch"
|
14 |
+
os.environ["STREAMLIT_CACHE_DIR"] = f"{CACHE_BASE}/streamlit_cache"
|
15 |
+
os.environ["STREAMLIT_STATIC_DIR"] = f"{CACHE_BASE}/streamlit_static"
|
16 |
+
|
17 |
+
# Create the directories before imports
|
18 |
+
for path in os.environ.values():
|
19 |
+
if path.startswith(CACHE_BASE):
|
20 |
+
os.makedirs(path, exist_ok=True)
|
21 |
+
|
22 |
+
# ====== Imports ======
|
23 |
+
import streamlit as st
|
24 |
import torch
|
|
|
|
|
25 |
from sentence_transformers import SentenceTransformer, util
|
26 |
+
from transformers import CLIPProcessor, CLIPModel
|
27 |
+
from datasets import load_dataset, get_dataset_split_names
|
28 |
+
from PIL import Image
|
29 |
+
import openai
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
|
31 |
+
# ========== π API Key ==========
|
32 |
+
openai.api_key = os.getenv("OPENAI_API_KEY")
|
|
|
33 |
|
34 |
+
# ========== π₯ Load Models ==========
|
35 |
+
@st.cache_resource(show_spinner=False)
|
36 |
+
def load_models():
|
37 |
+
clip_model = CLIPModel.from_pretrained(
|
38 |
+
"openai/clip-vit-base-patch32",
|
39 |
+
cache_dir=os.environ["TRANSFORMERS_CACHE"]
|
40 |
+
)
|
41 |
+
clip_processor = CLIPProcessor.from_pretrained(
|
42 |
+
"openai/clip-vit-base-patch32",
|
43 |
+
cache_dir=os.environ["TRANSFORMERS_CACHE"]
|
44 |
+
)
|
45 |
+
text_model = SentenceTransformer(
|
46 |
+
"all-MiniLM-L6-v2",
|
47 |
+
cache_folder=os.environ["SENTENCE_TRANSFORMERS_HOME"]
|
48 |
+
)
|
49 |
+
return clip_model, clip_processor, text_model
|
50 |
|
51 |
+
clip_model, clip_processor, text_model = load_models()
|
52 |
+
|
53 |
+
# ========== π₯ Load Dataset ==========
|
54 |
+
@st.cache_resource(show_spinner=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
|
56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
|
58 |
+
def load_medical_data():
|
59 |
+
available_splits = get_dataset_split_names("univanxx/3mdbench")
|
60 |
+
split_to_use = "train" if "train" in available_splits else available_splits[0]
|
61 |
+
dataset = load_dataset(
|
62 |
+
"univanxx/3mdbench",
|
63 |
+
split=split_to_use,
|
64 |
+
cache_dir=os.environ["HF_DATASETS_CACHE"]
|
65 |
+
)
|
66 |
+
return dataset
|
67 |
|
68 |
+
data = load_medical_data()
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
+
# Temporary debug display
|
71 |
+
#st.write("Dataset columns:", data.features.keys())
|
72 |
|
73 |
+
# After seeing the real column name, let's say it's "text" instead of "description":
|
74 |
+
text_field = "text" if "text" in data.features else list(data.features.keys())[0]
|
|
|
|
|
75 |
|
76 |
+
# Then use dynamic access:
|
77 |
+
#text_embeddings = embed_texts(data[text_field])
|
|
|
|
|
78 |
|
79 |
+
# ========== π§ Embedding Function ==========
|
80 |
+
@st.cache_data(show_spinner=False)
|
81 |
+
def embed_texts(_texts):
|
82 |
+
return text_model.encode(_texts, convert_to_tensor=True)
|
83 |
|
84 |
+
# Pick which text column to use
|
85 |
+
TEXT_COLUMN = "complaints" # or "general_complaint", depending on your needs
|
|
|
|
|
86 |
|
87 |
+
# ========== π§ββοΈ App UI ==========
|
88 |
+
st.title("π©Ί Multimodal Medical Chatbot")
|
89 |
|
90 |
+
query = st.text_input("Enter your medical question or symptom description:")
|
91 |
|
92 |
+
if query:
|
93 |
+
with st.spinner("Searching medical cases..."):
|
94 |
+
text_embeddings = embed_texts(data[TEXT_COLUMN])
|
95 |
+
query_embedding = embed_texts([query])[0]
|
96 |
|
97 |
+
# Compute similarity
|
98 |
+
cos_scores = util.pytorch_cos_sim(query_embedding, text_embeddings)[0]
|
99 |
+
top_result = torch.topk(cos_scores, k=1)
|
100 |
+
idx = top_result.indices[0].item()
|
101 |
+
selected = data[idx]
|
102 |
|
103 |
+
# Show Image
|
104 |
+
st.image(selected['image'], caption="Most relevant medical image", use_container_width=True)
|
|
|
|
|
105 |
|
106 |
+
# Show Text
|
107 |
+
st.markdown(f"**Case Description:** {selected[TEXT_COLUMN]}")
|
108 |
+
|
109 |
+
# GPT Explanation
|
110 |
+
if openai.api_key:
|
111 |
+
prompt = f"Explain this case in plain English: {selected[TEXT_COLUMN]}"
|
112 |
+
from openai import OpenAI
|
113 |
+
client = OpenAI(api_key=openai.api_key)
|
114 |
+
|
115 |
+
response = client.chat.completions.create(
|
116 |
+
model="gpt-4o", # or "gpt-4" if you need the older GPT-4
|
117 |
+
messages=[{"role": "user", "content": prompt}],
|
118 |
+
temperature=0.5,
|
119 |
+
max_tokens=150
|
120 |
+
)
|
121 |
+
explanation = response.choices[0].message.content
|
122 |
|
123 |
+
st.markdown(f"### π€ Explanation by GPT:\n{explanation}")
|
124 |
+
else:
|
125 |
+
st.warning("OpenAI API key not found. Please set OPENAI_API_KEY as a secret environment variable.")
|
126 |
|
127 |
+
st.caption("This chatbot is for educational purposes only and does not provide medical advice.")
|
|
|
|