File size: 6,121 Bytes
a794f66 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 |
import os
import json
import torch
import streamlit as st
import pandas as pd
from transformers import AutoModelForSequenceClassification, AutoTokenizer
# ==============================
# ⚙️ CONFIGURABLE PARAMETERS
# ==============================
MODEL_PATH = "dejanseo/bulgarian-search-query-intent-alpha" # HF model repository
LABEL_MAP_PATH = "label_map.json" # Ensure this file is in the same directory as app.py
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# ==============================
# 📌 Load Model and Tokenizer
# ==============================
@st.cache_resource
def load_inference_resources():
# Load the label mapping from local file
with open(LABEL_MAP_PATH, "r") as f:
label_map = json.load(f)
# Convert ID keys from string to int for id_to_label mapping
id_to_label = {int(k): v for k, v in label_map["id_to_label"].items()}
# Load the tokenizer and model from Hugging Face
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_PATH)
model.to(DEVICE)
model.eval() # Set model to evaluation mode
return model, tokenizer, label_map["label_to_id"], id_to_label
# ==============================
# 📌 Inference Function
# ==============================
def predict_intent(query, model, tokenizer, id_to_label):
"""
Predict the intent of a Bulgarian search query.
"""
# Tokenize input text
inputs = tokenizer(
query,
padding="max_length",
truncation=True,
max_length=128,
return_tensors="pt"
)
# Move inputs to the same device as the model
inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
# Inference without gradient tracking
with torch.no_grad():
outputs = model(**inputs)
# Compute probabilities with softmax
probabilities = torch.nn.functional.softmax(outputs.logits, dim=1)[0]
# Identify the predicted class and confidence
predicted_class_id = torch.argmax(probabilities).item()
predicted_intent = id_to_label[predicted_class_id]
confidence = probabilities[predicted_class_id].item()
# Build a dictionary with all intent scores
all_intents = {id_to_label[i]: prob.item() for i, prob in enumerate(probabilities)}
sorted_intents = sorted(all_intents.items(), key=lambda x: x[1], reverse=True)
return {
"query": query,
"predicted_intent": predicted_intent,
"confidence": confidence,
"all_scores": sorted_intents
}
# ==============================
# 🌟 Streamlit UI for Inference
# ==============================
def inference_ui():
st.title("🔍 Bulgarian Search Intent Classification")
try:
# Load resources
model, tokenizer, label_to_id, id_to_label = load_inference_resources()
st.success(f"✅ Model loaded successfully! Found {len(id_to_label)} intent classes.")
# Show available intents
with st.expander("Available Intent Classes"):
st.write(", ".join(id_to_label.values()))
# Single query inference
query = st.text_input("Enter a Bulgarian search query:", "Как да направя резервация за ресторант?")
if st.button("Predict Intent"):
with st.spinner("Analyzing query..."):
prediction = predict_intent(query, model, tokenizer, id_to_label)
st.subheader("Prediction Results")
st.metric(
label="Predicted Intent",
value=prediction["predicted_intent"],
delta=f"{prediction['confidence']*100:.2f}% confidence"
)
st.subheader("Intent Probabilities")
df_probs = pd.DataFrame(prediction["all_scores"], columns=["Intent", "Probability"])
df_top5 = df_probs.head(5)
st.bar_chart(df_top5.set_index("Intent"))
with st.expander("View All Intent Probabilities"):
st.dataframe(df_probs)
# Batch inference section
st.subheader("Batch Inference")
uploaded_file = st.file_uploader("Upload a CSV/Excel file with queries", type=["csv", "xlsx", "parquet"])
if uploaded_file is not None:
if uploaded_file.name.endswith(".csv"):
df = pd.read_csv(uploaded_file)
elif uploaded_file.name.endswith(".xlsx"):
df = pd.read_excel(uploaded_file)
elif uploaded_file.name.endswith(".parquet"):
df = pd.read_parquet(uploaded_file)
query_column = "query" if "query" in df.columns else st.selectbox("Select the column containing queries:", df.columns)
if query_column and st.button("Run Batch Inference"):
progress_bar = st.progress(0)
results = []
for i, row in enumerate(df[query_column]):
progress_bar.progress((i + 1) / len(df))
prediction = predict_intent(row, model, tokenizer, id_to_label)
results.append({
"query": row,
"predicted_intent": prediction["predicted_intent"],
"confidence": prediction["confidence"]
})
results_df = pd.DataFrame(results)
st.subheader("Batch Inference Results")
st.dataframe(results_df)
csv = results_df.to_csv(index=False)
st.download_button(
label="Download Results as CSV",
data=csv,
file_name="batch_inference_results.csv",
mime="text/csv"
)
except Exception as e:
st.error(f"❌ Error loading model: {str(e)}")
st.error("Please ensure the model and label map files are available.")
if __name__ == "__main__":
inference_ui()
|