Suramya/Llama-2-7b-CloudLex-Intent-Detection

Llama-2-7B CloudLex Intent Detection (QLoRA)

Overview

This model is a CloudLex-specific intent classification model fine-tuned from
NousResearch/Llama-2-7b-chat-hf using QLoRA (4-bit) and LoRA adapters.

It is designed to classify user messages into predefined business intents commonly encountered in legal-tech and SaaS customer interactions.

---

🎯 Task

Single-label intent classification (SEQ_CLS)

Given a user message, the model predicts one intent from the following set:

• Buying
• Support
• Careers
• Partnership
• Explore
• Others

---

🧠 Training Details

• Base model: NousResearch/Llama-2-7b-chat-hf
• Fine-tuning method: QLoRA (PEFT)
• Quantization: 4-bit NF4
• LoRA rank: 64
• LoRA alpha: 16
• Optimizer: paged_adamw_32bit
• Training data: ~1,000 balanced intent-labeled samples
• Framework: Hugging Face Transformers + PEFT

This repository contains LoRA adapters only, not the full base model.

---

🚀 How to Use (Inference)

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
from peft import PeftModel

# Load base model
base_model = AutoModelForSequenceClassification.from_pretrained(
    "NousResearch/Llama-2-7b-chat-hf",
    load_in_4bit=True,
    device_map="auto"
)

# Load LoRA adapters
model = PeftModel.from_pretrained(
    base_model,
    "Suramya/Llama-2-7b-CloudLex-Intent-Detection"
)

tokenizer = AutoTokenizer.from_pretrained(
    "Suramya/Llama-2-7b-CloudLex-Intent-Detection"
)

classifier = pipeline(
    "text-classification",
    model=model,
    tokenizer=tokenizer
)

classifier("I'd like to schedule a demo for our law firm")