Spaces:
Sleeping
Sleeping
refactor: improve response
Browse files- app.py +48 -34
- requirements.txt +1 -1
app.py
CHANGED
|
@@ -1,10 +1,9 @@
|
|
| 1 |
import streamlit as st
|
| 2 |
-
|
| 3 |
-
import os
|
| 4 |
-
from datasets import load_from_disk, Dataset
|
| 5 |
import torch
|
| 6 |
import logging
|
| 7 |
-
import
|
|
|
|
| 8 |
import arxiv
|
| 9 |
import requests
|
| 10 |
import xml.etree.ElementTree as ET
|
|
@@ -17,14 +16,14 @@ logging.basicConfig(level=logging.INFO)
|
|
| 17 |
DATA_DIR = "/data" if os.path.exists("/data") else "."
|
| 18 |
DATASET_DIR = os.path.join(DATA_DIR, "rag_dataset")
|
| 19 |
DATASET_PATH = os.path.join(DATASET_DIR, "dataset")
|
| 20 |
-
MODEL_PATH = "t5-small" #
|
| 21 |
|
| 22 |
@st.cache_resource
|
| 23 |
def load_local_model():
|
| 24 |
"""Load the local Hugging Face model"""
|
| 25 |
try:
|
| 26 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
|
| 27 |
-
model =
|
| 28 |
MODEL_PATH,
|
| 29 |
device_map={"": "cpu"}, # Force CPU
|
| 30 |
torch_dtype=torch.float32
|
|
@@ -206,37 +205,46 @@ def generate_answer(question, context, max_length=512):
|
|
| 206 |
|
| 207 |
# Clean and format the context
|
| 208 |
clean_context = clean_text(context)
|
|
|
|
| 209 |
|
| 210 |
-
# Format the
|
| 211 |
-
|
| 212 |
|
| 213 |
-
Question: {
|
| 214 |
|
| 215 |
Research Papers:
|
| 216 |
{clean_context}
|
| 217 |
|
| 218 |
-
Instructions:
|
| 219 |
-
1.
|
| 220 |
-
2.
|
| 221 |
-
3.
|
| 222 |
-
4. Notes any limitations
|
| 223 |
|
| 224 |
-
|
| 225 |
|
| 226 |
try:
|
| 227 |
-
#
|
| 228 |
-
inputs = tokenizer(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 229 |
|
| 230 |
with torch.inference_mode():
|
| 231 |
outputs = model.generate(
|
| 232 |
**inputs,
|
| 233 |
max_length=max_length,
|
| 234 |
-
min_length=
|
| 235 |
-
num_beams=
|
| 236 |
length_penalty=1.5,
|
| 237 |
temperature=0.7,
|
| 238 |
repetition_penalty=1.2,
|
| 239 |
-
early_stopping=True
|
|
|
|
|
|
|
|
|
|
|
|
|
| 240 |
)
|
| 241 |
|
| 242 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
@@ -247,28 +255,34 @@ Keep your answer focused, clear, and helpful for someone wanting to understand a
|
|
| 247 |
return f"""Here's what we know about autism in relation to your question:
|
| 248 |
|
| 249 |
1. General Understanding:
|
| 250 |
-
- Autism Spectrum Disorder (ASD) is a complex
|
| 251 |
-
- It affects how a person
|
| 252 |
-
- Each
|
|
|
|
| 253 |
|
| 254 |
-
2. Key
|
| 255 |
-
-
|
| 256 |
- Repetitive behaviors and specific interests
|
| 257 |
- Sensory sensitivities
|
| 258 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 259 |
|
| 260 |
-
|
| 261 |
-
-
|
| 262 |
-
|
| 263 |
-
|
| 264 |
-
|
| 265 |
-
* Effective interventions and supports
|
| 266 |
|
| 267 |
For more specific information, try asking about:
|
| 268 |
-
- Specific
|
| 269 |
- Diagnostic processes
|
| 270 |
- Treatment approaches
|
| 271 |
-
-
|
| 272 |
|
| 273 |
# Format the response for better readability
|
| 274 |
formatted_response = response.replace(". ", ".\n").replace("• ", "\n• ")
|
|
|
|
| 1 |
import streamlit as st
|
| 2 |
+
import pandas as pd
|
|
|
|
|
|
|
| 3 |
import torch
|
| 4 |
import logging
|
| 5 |
+
import os
|
| 6 |
+
from transformers import AutoTokenizer, T5ForConditionalGeneration
|
| 7 |
import arxiv
|
| 8 |
import requests
|
| 9 |
import xml.etree.ElementTree as ET
|
|
|
|
| 16 |
DATA_DIR = "/data" if os.path.exists("/data") else "."
|
| 17 |
DATASET_DIR = os.path.join(DATA_DIR, "rag_dataset")
|
| 18 |
DATASET_PATH = os.path.join(DATASET_DIR, "dataset")
|
| 19 |
+
MODEL_PATH = "google/flan-t5-small" # Using flan-t5-small for better performance
|
| 20 |
|
| 21 |
@st.cache_resource
|
| 22 |
def load_local_model():
|
| 23 |
"""Load the local Hugging Face model"""
|
| 24 |
try:
|
| 25 |
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
|
| 26 |
+
model = T5ForConditionalGeneration.from_pretrained(
|
| 27 |
MODEL_PATH,
|
| 28 |
device_map={"": "cpu"}, # Force CPU
|
| 29 |
torch_dtype=torch.float32
|
|
|
|
| 205 |
|
| 206 |
# Clean and format the context
|
| 207 |
clean_context = clean_text(context)
|
| 208 |
+
clean_question = clean_text(question)
|
| 209 |
|
| 210 |
+
# Format the input for T5 (it expects a specific format)
|
| 211 |
+
input_text = f"""Answer the following question about autism using the provided research papers.
|
| 212 |
|
| 213 |
+
Question: {clean_question}
|
| 214 |
|
| 215 |
Research Papers:
|
| 216 |
{clean_context}
|
| 217 |
|
| 218 |
+
Instructions: Provide a detailed answer that:
|
| 219 |
+
1. Explains the main concepts clearly
|
| 220 |
+
2. Uses specific evidence from the research
|
| 221 |
+
3. Discusses practical implications
|
| 222 |
+
4. Notes any limitations
|
| 223 |
|
| 224 |
+
Answer:"""
|
| 225 |
|
| 226 |
try:
|
| 227 |
+
# T5 expects a specific format for the input
|
| 228 |
+
inputs = tokenizer(input_text,
|
| 229 |
+
return_tensors="pt",
|
| 230 |
+
max_length=1024,
|
| 231 |
+
truncation=True,
|
| 232 |
+
padding=True)
|
| 233 |
|
| 234 |
with torch.inference_mode():
|
| 235 |
outputs = model.generate(
|
| 236 |
**inputs,
|
| 237 |
max_length=max_length,
|
| 238 |
+
min_length=100,
|
| 239 |
+
num_beams=5,
|
| 240 |
length_penalty=1.5,
|
| 241 |
temperature=0.7,
|
| 242 |
repetition_penalty=1.2,
|
| 243 |
+
early_stopping=True,
|
| 244 |
+
no_repeat_ngram_size=3,
|
| 245 |
+
do_sample=True,
|
| 246 |
+
top_k=50,
|
| 247 |
+
top_p=0.95
|
| 248 |
)
|
| 249 |
|
| 250 |
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
|
|
|
| 255 |
return f"""Here's what we know about autism in relation to your question:
|
| 256 |
|
| 257 |
1. General Understanding:
|
| 258 |
+
- Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition
|
| 259 |
+
- It affects how a person perceives, communicates, and interacts with the world
|
| 260 |
+
- Each individual with autism has unique strengths and challenges
|
| 261 |
+
- Early identification and support are crucial
|
| 262 |
|
| 263 |
+
2. Key Characteristics:
|
| 264 |
+
- Social communication and interaction patterns
|
| 265 |
- Repetitive behaviors and specific interests
|
| 266 |
- Sensory sensitivities
|
| 267 |
+
- Variable cognitive and language abilities
|
| 268 |
+
|
| 269 |
+
3. Important Considerations:
|
| 270 |
+
- Autism is a spectrum, meaning it affects each person differently
|
| 271 |
+
- Support needs vary from person to person
|
| 272 |
+
- Many individuals with autism have unique talents and abilities
|
| 273 |
+
- Research continues to improve our understanding
|
| 274 |
|
| 275 |
+
4. Current Research Areas:
|
| 276 |
+
- Brain development and neurology
|
| 277 |
+
- Genetic and environmental factors
|
| 278 |
+
- Early intervention methods
|
| 279 |
+
- Support strategies and therapies
|
|
|
|
| 280 |
|
| 281 |
For more specific information, try asking about:
|
| 282 |
+
- Specific autism characteristics
|
| 283 |
- Diagnostic processes
|
| 284 |
- Treatment approaches
|
| 285 |
+
- Latest research findings"""
|
| 286 |
|
| 287 |
# Format the response for better readability
|
| 288 |
formatted_response = response.replace(". ", ".\n").replace("• ", "\n• ")
|
requirements.txt
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
streamlit>=1.32.0
|
| 2 |
-
transformers
|
| 3 |
datasets>=2.17.0
|
| 4 |
--extra-index-url https://download.pytorch.org/whl/cpu
|
| 5 |
torch>=2.2.0
|
|
|
|
| 1 |
streamlit>=1.32.0
|
| 2 |
+
transformers==4.36.2
|
| 3 |
datasets>=2.17.0
|
| 4 |
--extra-index-url https://download.pytorch.org/whl/cpu
|
| 5 |
torch>=2.2.0
|