ModelHubManager / components /model_inference.py
S-Dreamer's picture
Upload 31 files
74dd3f1 verified
import streamlit as st
import json
import pandas as pd
import numpy as np
import plotly.express as px
from io import StringIO
import time
def model_inference_dashboard(model_info):
"""Create a dashboard for testing model inference directly in the app"""
if not model_info:
st.error("Model information not found")
return
st.subheader("🧠 Model Inference Dashboard")
# Get the pipeline type based on model tags or information
pipeline_tag = getattr(model_info, "pipeline_tag", None)
if not pipeline_tag:
# Try to determine from tags
tags = getattr(model_info, "tags", [])
for tag in tags:
if tag in [
"text-classification", "token-classification", "question-answering",
"summarization", "translation", "text-generation", "fill-mask",
"sentence-similarity", "image-classification", "object-detection",
"image-segmentation", "text-to-image", "image-to-text"
]:
pipeline_tag = tag
break
if not pipeline_tag:
pipeline_tag = "text-classification" # Default fallback
# Display information about the model
st.info(f"This dashboard allows you to test your model's inference capabilities. Model pipeline: **{pipeline_tag}**")
# Different input options based on pipeline type
input_data = None
if pipeline_tag in ["text-classification", "token-classification", "fill-mask", "text-generation", "summarization"]:
# Text-based input
st.markdown("### Text Input")
input_text = st.text_area(
"Enter text for inference",
value="This model is amazing!",
height=150
)
# Additional parameters for specific pipelines
if pipeline_tag == "text-generation":
col1, col2 = st.columns(2)
with col1:
max_length = st.slider("Max Length", min_value=10, max_value=500, value=100)
with col2:
temperature = st.slider("Temperature", min_value=0.1, max_value=2.0, value=1.0, step=0.1)
input_data = {
"text": input_text,
"max_length": max_length,
"temperature": temperature
}
elif pipeline_tag == "summarization":
max_length = st.slider("Max Summary Length", min_value=10, max_value=200, value=50)
input_data = {
"text": input_text,
"max_length": max_length
}
else:
input_data = {"text": input_text}
elif pipeline_tag in ["question-answering"]:
st.markdown("### Question & Context")
question = st.text_input("Question", value="What is this model about?")
context = st.text_area(
"Context",
value="This model is a transformer-based language model designed for natural language understanding tasks.",
height=150
)
input_data = {
"question": question,
"context": context
}
elif pipeline_tag in ["translation"]:
st.markdown("### Translation")
source_lang = st.selectbox("Source Language", ["English", "French", "German", "Spanish", "Chinese"])
target_lang = st.selectbox("Target Language", ["French", "English", "German", "Spanish", "Chinese"])
translation_text = st.text_area("Text to translate", value="Hello, how are you?", height=150)
input_data = {
"text": translation_text,
"source_language": source_lang,
"target_language": target_lang
}
elif pipeline_tag in ["image-classification", "object-detection", "image-segmentation"]:
st.markdown("### Image Input")
upload_method = st.radio("Select input method", ["Upload Image", "Image URL"])
if upload_method == "Upload Image":
uploaded_file = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])
if uploaded_file is not None:
st.image(uploaded_file, caption="Uploaded Image", use_column_width=True)
input_data = {"image": uploaded_file}
else:
image_url = st.text_input("Image URL", value="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/distilbert-base-uncased-finetuned-sst-2-english-architecture.png")
if image_url:
st.image(image_url, caption="Image from URL", use_column_width=True)
input_data = {"image_url": image_url}
elif pipeline_tag in ["audio-classification", "automatic-speech-recognition"]:
st.markdown("### Audio Input")
upload_method = st.radio("Select input method", ["Upload Audio", "Audio URL"])
if upload_method == "Upload Audio":
uploaded_file = st.file_uploader("Upload an audio file", type=["mp3", "wav", "ogg"])
if uploaded_file is not None:
st.audio(uploaded_file)
input_data = {"audio": uploaded_file}
else:
audio_url = st.text_input("Audio URL")
if audio_url:
st.audio(audio_url)
input_data = {"audio_url": audio_url}
# Execute inference
if st.button("Run Inference", use_container_width=True):
if input_data:
with st.spinner("Running inference..."):
# In a real implementation, this would call the HF Inference API
# For demo purposes, simulate a response
time.sleep(2)
# Generate a sample response based on the pipeline type
if pipeline_tag == "text-classification":
result = [
{"label": "POSITIVE", "score": 0.9231},
{"label": "NEGATIVE", "score": 0.0769}
]
elif pipeline_tag == "token-classification":
result = [
{"entity": "B-PER", "word": "This", "score": 0.2, "index": 0, "start": 0, "end": 4},
{"entity": "O", "word": "model", "score": 0.95, "index": 1, "start": 5, "end": 10},
{"entity": "O", "word": "is", "score": 0.99, "index": 2, "start": 11, "end": 13},
{"entity": "B-MISC", "word": "amazing", "score": 0.85, "index": 3, "start": 14, "end": 21}
]
elif pipeline_tag == "text-generation":
result = {
"generated_text": input_data["text"] + " It provides state-of-the-art performance on a wide range of natural language processing tasks, including sentiment analysis, named entity recognition, and question answering. The model was trained on a diverse corpus of text data, allowing it to generate coherent and contextually relevant responses."
}
elif pipeline_tag == "summarization":
result = {
"summary_text": "This model provides excellent performance."
}
elif pipeline_tag == "question-answering":
result = {
"answer": "a transformer-based language model",
"start": 9,
"end": 45,
"score": 0.953
}
elif pipeline_tag == "translation":
if input_data["target_language"] == "French":
result = {"translation_text": "Bonjour, comment allez-vous?"}
elif input_data["target_language"] == "German":
result = {"translation_text": "Hallo, wie geht es dir?"}
elif input_data["target_language"] == "Spanish":
result = {"translation_text": "Hola, ¿cómo estás?"}
elif input_data["target_language"] == "Chinese":
result = {"translation_text": "你好,你好吗?"}
else:
result = {"translation_text": "Hello, how are you?"}
elif pipeline_tag in ["image-classification"]:
result = [
{"label": "diagram", "score": 0.9712},
{"label": "architecture", "score": 0.0231},
{"label": "document", "score": 0.0057}
]
elif pipeline_tag in ["object-detection"]:
result = [
{"label": "box", "score": 0.9712, "box": {"xmin": 10, "ymin": 20, "xmax": 100, "ymax": 80}},
{"label": "text", "score": 0.8923, "box": {"xmin": 120, "ymin": 30, "xmax": 250, "ymax": 60}}
]
else:
result = {"result": "Sample response for " + pipeline_tag}
# Display the results
st.markdown("### Inference Results")
# Different visualizations based on the response type
if pipeline_tag == "text-classification":
# Create a bar chart for classification results
result_df = pd.DataFrame(result)
fig = px.bar(
result_df,
x="label",
y="score",
color="score",
color_continuous_scale=px.colors.sequential.Viridis,
title="Classification Results"
)
st.plotly_chart(fig, use_container_width=True)
# Show the raw results
st.json(result)
elif pipeline_tag == "token-classification":
# Display entity highlighting
st.markdown("#### Named Entities")
# Create HTML with colored spans for entities
html = ""
input_text = input_data["text"]
entities = {}
for item in result:
if item["entity"].startswith("B-") or item["entity"].startswith("I-"):
entity_type = item["entity"][2:] # Remove B- or I- prefix
entities[entity_type] = entities.get(entity_type, 0) + 1
# Create a color map for entity types
colors = px.colors.qualitative.Plotly[:len(entities)]
entity_colors = dict(zip(entities.keys(), colors))
# Create the HTML
for item in result:
word = item["word"]
entity = item["entity"]
if entity == "O":
html += f"{word} "
else:
entity_type = entity[2:] if entity.startswith("B-") or entity.startswith("I-") else entity
color = entity_colors.get(entity_type, "#CCCCCC")
html += f'<span style="background-color: {color}; padding: 2px; border-radius: 3px;" title="{entity} ({item["score"]:.2f})">{word}</span> '
st.markdown(f'<div style="line-height: 2.5;">{html}</div>', unsafe_allow_html=True)
# Display legend
st.markdown("#### Entity Legend")
legend_html = "".join([
f'<span style="background-color: {color}; padding: 2px 8px; margin-right: 10px; border-radius: 3px;">{entity}</span>'
for entity, color in entity_colors.items()
])
st.markdown(f'<div>{legend_html}</div>', unsafe_allow_html=True)
# Show the raw results
st.json(result)
elif pipeline_tag in ["text-generation", "summarization", "translation"]:
# Display the generated text
response_key = "generated_text" if "generated_text" in result else "summary_text" if "summary_text" in result else "translation_text"
st.markdown(f"#### Output Text")
st.markdown(f'<div style="background-color: #f0f2f6; padding: 20px; border-radius: 10px;">{result[response_key]}</div>', unsafe_allow_html=True)
# Text stats
st.markdown("#### Text Statistics")
input_length = len(input_data["text"]) if "text" in input_data else 0
output_length = len(result[response_key])
col1, col2, col3 = st.columns(3)
with col1:
st.metric("Input Length", input_length, "characters")
with col2:
st.metric("Output Length", output_length, "characters")
with col3:
compression = ((output_length - input_length) / input_length * 100) if input_length > 0 else 0
st.metric("Length Change", f"{compression:.1f}%", f"{output_length - input_length} chars")
elif pipeline_tag == "question-answering":
# Highlight the answer in the context
st.markdown("#### Answer")
st.markdown(f'<div style="background-color: #e6f3ff; padding: 10px; border-radius: 5px; font-weight: bold;">{result["answer"]}</div>', unsafe_allow_html=True)
# Show the answer in context
if "context" in input_data:
st.markdown("#### Answer in Context")
context = input_data["context"]
start = result["start"]
end = result["end"]
highlighted_context = (
context[:start] +
f'<span style="background-color: #ffeb3b; font-weight: bold;">{context[start:end]}</span>' +
context[end:]
)
st.markdown(f'<div style="background-color: #f0f2f6; padding: 15px; border-radius: 10px; line-height: 1.5;">{highlighted_context}</div>', unsafe_allow_html=True)
# Confidence score
st.markdown("#### Confidence")
st.progress(result["score"])
st.text(f"Confidence Score: {result['score']:.4f}")
elif pipeline_tag == "image-classification":
# Create a bar chart for classification results
result_df = pd.DataFrame(result)
fig = px.bar(
result_df,
x="score",
y="label",
orientation='h',
color="score",
color_continuous_scale=px.colors.sequential.Viridis,
title="Image Classification Results"
)
fig.update_layout(yaxis={'categoryorder':'total ascending'})
st.plotly_chart(fig, use_container_width=True)
# Show the raw results
st.json(result)
else:
# Generic display for other types
st.json(result)
# Option to save the results
st.download_button(
label="Download Results",
data=json.dumps(result, indent=2),
file_name="inference_results.json",
mime="application/json"
)
else:
st.warning("Please provide input data for inference")
# API integration options
with st.expander("API Integration"):
st.markdown("### Use this model in your application")
# Python code example
st.markdown("#### Python")
python_code = f"""
```python
import requests
API_URL = "https://api-inference.huggingface.co/models/{model_info.modelId}"
headers = {{"Authorization": "Bearer YOUR_API_KEY"}}
def query(payload):
response = requests.post(API_URL, headers=headers, json=payload)
return response.json()
# Example usage
output = query({{
"inputs": "This model is amazing!"
}})
print(output)
```
"""
st.markdown(python_code)
# JavaScript code example
st.markdown("#### JavaScript")
js_code = f"""
```javascript
async function query(data) {{
const response = await fetch(
"https://api-inference.huggingface.co/models/{model_info.modelId}",
{{
headers: {{ Authorization: "Bearer YOUR_API_KEY" }},
method: "POST",
body: JSON.stringify(data),
}}
);
const result = await response.json();
return result;
}}
// Example usage
query({{"inputs": "This model is amazing!"}}).then((response) => {{
console.log(JSON.stringify(response));
}});
```
"""
st.markdown(js_code)