space add

.env

@@ -0,0 +1,2 @@
+GLICLASS_MODEL_PATH=knowledgator/gliclass_msmarco_merged
+MAX_DOCS=8

app.py

@@ -0,0 +1,13 @@
+from dotenv import load_dotenv
+import gradio as gr
+load_dotenv()
+
+from interfaces import landing_interface, main_pipeline
+
+demo = gr.TabbedInterface([landing_interface, main_pipeline],
+                          ["Introduction", "Reranking"],
+                          title="GLiClass SandBox",
+                          theme=gr.themes.Base())
+
+demo.queue()
+demo.launch(debug=True, share=True)

interfaces/__init__.py

@@ -0,0 +1,2 @@
+from .landing import landing_interface
+from .main_pipeline import main_pipeline

interfaces/landing.py

@@ -0,0 +1,41 @@
+import gradio as gr
+
+
+with open('materials/introduction.html', 'r', encoding='utf-8') as file:
+    html_description = file.read()
+
+with gr.Blocks() as landing_interface:
+    gr.HTML(html_description)
+    
+    with gr.Accordion("How to run this model locally", open=False):
+        gr.Markdown(
+            """
+            ## Installation
+            To use this model, you must install the GLiClass Python library:
+            ```
+            !pip install gliclass
+            ```
+         
+            ## Usage
+            Once you've downloaded the GLiClass library, you can import the GLiClassModel and ZeroShotClassificationPipeline classes.
+            """
+        )
+        gr.Code(
+            '''
+from gliclass import GLiClassModel, ZeroShotClassificationPipeline
+from transformers import AutoTokenizer
+
+model = GLiClassModel.from_pretrained("knowledgator/gliclass-small-v1")
+tokenizer = AutoTokenizer.from_pretrained("knowledgator/gliclass-small-v1")
+
+pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device='cuda:0')
+
+text = "One day I will see the world!"
+labels = ["travel", "dreams", "sport", "science", "politics"]
+results = pipeline(text, labels, threshold=0.5)[0] #because we have one text
+
+for result in results:
+    print(result["label"], "=>", result["score"])
+            ''',
+            language="python",
+        )

interfaces/main_pipeline.py

@@ -0,0 +1,132 @@
+import os
+import torch
+import gradio as gr
+from typing import List
+from transformers import AutoTokenizer
+from gliclass import GLiClassModel, ZeroShotClassificationPipeline
+
+
+device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
+
+
+model = GLiClassModel.from_pretrained(os.getenv("GLICLASS_MODEL_PATH")).eval().to(device)
+tokenizer = AutoTokenizer.from_pretrained(os.getenv("GLICLASS_MODEL_PATH"))
+multi_label_pipeline = ZeroShotClassificationPipeline(model, tokenizer, classification_type='multi-label', device=device)
+
+
+example_1 =[
+    "I want to live in New York.",
+    'York is a cathedral city in North Yorkshire, England, with Roman origins',
+    'San Francisco,[23] officially the City and County of San Francisco, is a commercial, financial, and cultural center within Northern California, United States.',
+    'New York, often called New York City (NYC),[b] is the most populous city in the United States',
+    "New York City is the third album by electronica group Brazilian Girls, released in 2008.",
+    "New York City was an American R&B vocal group.",
+    "New York City is an album by the Peter Malick Group featuring Norah Jones.",
+    "New York City: The Album is the debut studio album by American rapper Troy Ave. ",
+    '"New York City" is a song by British new wave band The Armoury Show',
+]
+
+example_2 = [
+    "Looking for waterproof hiking boots that can handle freezing temperatures and rugged terrain.",
+    "TrailMaster X200 – waterproof boots with Vibram Arctic Grip soles, rated for -20°C and rocky paths.",
+    "UrbanStep Sneakers – stylish and breathable, not designed for rugged use or cold weather.",
+    "AlpineShield GTX – Gore-Tex lining, insulated to -15°C, ideal for mountain hiking.",
+    "Desert Trek Sandals – open-toe design, breathable and lightweight, not waterproof.",
+    "SummitPro Winter Boots – fleece-lined, waterproof up to ankle depth, tested to -5°C.",
+    "Marathon Lite – road-running shoes with shock-absorbing soles, non-waterproof.",
+    "TrailMaster X100 – waterproof boots with basic insulation, effective down to 0°C.",
+    "Climber Pro GTX – reinforced toe cap, Gore-Tex membrane, insulated to -20°C, certified for alpine routes."
+]
+
+example_3 = [
+    "Our users are reporting 504 Gateway Timeout errors when accessing the app during peak hours.",
+    "A 504 Gateway Timeout indicates that a server did not receive a timely response from another server upstream.",
+    "A 502 Bad Gateway occurs when the server, acting as a gateway, receives an invalid response from the upstream server.",
+    "Common causes of 504 errors include high server load, network congestion, or misconfigured backend timeouts.",
+    "A 403 Forbidden error suggests that the server is refusing to authorize the request, often due to permissions.",
+    "To resolve 504 errors, check server logs, backend service availability, and increase timeout settings if necessary.",
+    "A 408 Request Timeout is returned when the client fails to send a complete request in time.",
+    "A 500 Internal Server Error is a generic error indicating that the server encountered an unexpected condition.",
+    "Network latency monitoring tools can help identify bottlenecks that may cause 504 errors during high traffic periods."
+]
+
+example_4 = [
+    "A 45-year-old male presents with persistent cough, night sweats, low-grade fever, and weight loss over 3 months.",
+    "Lung cancer can cause cough and weight loss; however, it often includes hemoptysis and may show a solitary mass on imaging.",
+    "Bronchiectasis is characterized by chronic productive cough and recurrent infections but usually lacks significant weight loss.",
+    "Pneumonia presents acutely with high fever, productive cough, and may show lobar consolidation on imaging.",
+    "Sarcoidosis may cause cough and weight loss, with bilateral hilar lymphadenopathy seen on chest X-ray.",
+    "Tuberculosis typically presents with chronic cough, night sweats, weight loss, and may show upper lobe infiltrates on chest X-ray.",
+    "Chronic obstructive pulmonary disease (COPD) often involves chronic cough and dyspnea but is less associated with night sweats.",
+    "Fungal lung infections like histoplasmosis can mimic TB symptoms but are more common in specific endemic regions.",
+    "Gastroesophageal reflux disease (GERD) can cause chronic cough, but without systemic symptoms like weight loss or fever."
+]
+
+example_5 = [
+    "How can I set up a recurring payment for my monthly rent via online banking?",
+    "A standing order allows you to set up automatic fixed-amount payments on a regular schedule (e.g., monthly rent) through your bank.",
+    "A direct debit authorizes a third party to withdraw variable amounts from your account, typically used for utility bills.",
+    "Wire transfers are typically one-off payments that do not recur automatically.",
+    "You can schedule a one-time payment for a future date using the online banking portal, but it won’t repeat monthly.",
+    "Bank-issued cashier’s checks are used for large payments but require manual setup each time.",
+    "To set up recurring credit card payments, navigate to your card provider’s auto-pay settings (note: for card bills only).",
+    "Standing orders can be modified or canceled at any time via your online banking dashboard.",
+    "International transfers may incur additional fees and are not ideal for domestic rent payments."
+]
+
+examples = [
+    example + [""] * (int(os.getenv("MAX_DOCS")) - len(example) -1) for example in [example_1, example_2, example_3, example_4, example_5]
+]
+
+
+def classification(*args) -> List[str]:
+    labels = [arg for arg in args[1:]]
+    labels = list(filter(None, labels))
+    query = args[0]
+    
+    results = sorted(multi_label_pipeline(query, labels, threshold=0.0)[0], key=lambda x: x["score"], reverse=True)
+    docs = []
+    scores = []
+    for predict in results:
+        docs.append(predict["label"])
+        scores.append(round(predict["score"], 2))
+    for _ in range(int(os.getenv("MAX_DOCS")) - len(docs)):
+        docs.append("")
+        scores.append("")
+    return docs + scores
+
+with gr.Blocks(title="GLiClass-Reranker") as main_pipeline:
+    inputs = []
+    outputs = []
+    query = gr.Textbox(
+        value=examples[0][0], label="Text query", placeholder="Enter your query here", lines=10
+    )
+    submit_btn = gr.Button("Rerank")
+    inputs.append(query)
+    for i in range(int(os.getenv("MAX_DOCS"))):
+        with gr.Group():
+            doc_input = gr.Textbox(
+                value=examples[0][1+i],
+                label=f"Document {i}",
+                placeholder="Enter your labels here (comma separated)",
+                scale=2,
+            )
+            score_output = gr.Textbox(
+                label=f"Score {i}",
+                placeholder="Score will appear here",
+                scale=2,
+            )
+            inputs.append(doc_input)
+            outputs.append(score_output)
+    outputs = inputs[1:] + outputs
+    examples = gr.Examples(
+        examples=examples,
+        fn=classification,
+        inputs=inputs,
+        outputs=outputs,
+        cache_examples=True,
+    )
+
+    submit_btn.click(
+        fn=classification, inputs=inputs, outputs=outputs
+    )

materials/introduction.html

@@ -0,0 +1,67 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>GLiClass Reranker</title>
+    <style>
+        body {
+            font-family: Arial, sans-serif;
+            margin: 10px;
+            line-height: 1.6;
+        }
+        .link-button {
+            display: inline-flex;
+            align-items: center;
+            justify-content: center;
+            margin: 10px;
+            padding: 10px;
+            background-color: white;
+            border: 1px solid grey;
+            color: #007bff;
+            text-decoration: none;
+            border-radius: 10px;
+            text-align: center;
+            vertical-align: middle;
+            box-sizing: border-box;
+        }
+        .link-button:hover {
+            background-color: #c0dcfc;
+        }
+        .link-button img {
+            height: 20px;
+            width: auto;
+            display: block;
+        }
+        .links-container {
+            text-align: center;
+            margin: auto;
+            display: flex;
+            justify-content: center;
+            flex-wrap: wrap;
+        }
+    </style>
+</head>
+<body>
+    <h2>Welcome to GliClass Reranker!</h2>
+    <h3>GLiClass: Generalist and Lightweight Model for Sequence Classification:</h3>
+    <ol>
+        <li><b>Reranker in RAG pipelines</b></li>
+        <li><b>Zero-Shot text classification</b></li>
+        <li><b>Topic Classification</b></li>
+        <li><b>Sentiment Analysis</b></li>
+    </ol>
+    <h3>What is GliClass?</h3>
+    <p>This is an efficient zero-shot classifier inspired by <a href="https://github.com/urchade/GLiNER/tree/main">"GLiNER"</a> work. It demonstrates the same performance as a cross-encoder while being more compute-efficient because classification is done at a single forward path. It can be used for topic classification, sentiment analysis and as a reranker in RAG pipelines.<p>
+    <h3>Remember, information extraction is not just about data; it's about insights. Let's uncover those insights together!💫</h3>
+    <!-- Links Section -->
+    <div class="links-container">
+        <a href="https://www.knowledgator.com/" class="link-button" target="_blank"><img src="https://assets-global.website-files.com/65902be8ba48a05dfdb73331/6590476fcc8e8f35b2332781_Group%201000002504%20(1).png" alt="Visit our website"></a>
+        <a href="https://www.linkedin.com/company/knowledgator/" class="link-button" target="_blank"><img src="https://upload.wikimedia.org/wikipedia/commons/thumb/0/01/LinkedIn_Logo.svg/1280px-LinkedIn_Logo.svg.png" alt="Follow on LinkedIn"></a>
+        <a href="https://huggingface.co/knowledgator/" class="link-button" target="_blank"><img src="https://vectorseek.com/wp-content/uploads/2023/12/Hugging-Face-Logo-Vector.svg-.png" alt="Hugging Face Profile"></a>
+        <a href="https://twitter.com/knowledgator" class="link-button" target="_blank"><img src="https://freepnglogo.com/images/all_img/1691832278twitter-x-logo-png.png" alt="Follow on X"></a>
+        <a href="https://blog.knowledgator.com/" class="link-button" target="_blank"><img src="https://upload.wikimedia.org/wikipedia/commons/thumb/0/0d/Medium_%28website%29_logo.svg/2560px-Medium_%28website%29_logo.svg.png" alt="Follow on Medium"></a>
+        <a href="https://discord.com/invite/dkyeAgs9DG" class="link-button" target="_blank"><img src="https://assets-global.website-files.com/6257adef93867e50d84d30e2/636e0b52aa9e99b832574a53_full_logo_blurple_RGB.png" alt="Join our Discord"></a>
+    </div>
+</body>
+</html>

requirements.txt

@@ -0,0 +1,3 @@
+gliclass
+gradio
+dotenv