---
---
language:
  - en
base_model:
  - meta-llama/Llama-3.3-70B-Instruct-Reference
license: other
tags:
  - human-resources
  - hr-management
  - question-answering
  - text-generation
  - adaptive-data
  - autosscientist
  - lora
pipeline_tag: text-generation
credit: "Adaptive Data by Adaption Labs"
---

# HR Management QA — Adaptive Data

> A focused Human Resources dataset designed to improve AI responses to practical HR questions and workplace scenarios.

## Overview

This dataset contains **113 HR-focused examples** covering recruitment, employee development, workplace policies, attendance, maternity leave, performance management, career development, and employee support.

The dataset was enhanced using **Adaption Labs Adaptive Data** and evaluated through **AutoScientist**.

## Dataset Summary

| Feature | Details |
|---|---|
| Domain | Human Resources |
| Language | English |
| Examples | 113 |
| Format | Question & Answer |
| Task | HR Question Answering |
| Base Model | `meta-llama/Llama-3.3-70B-Instruct-Reference` |
| Fine-tuning | LoRA |

## Adaptive Data Results

| Metric | Before | After |
|---|---:|---:|
| Quality Score | 5.0 | **7.7** |
| Custom Rubric | 5.8 | **7.2** |
| Grade | C | **B** |
| Percentile | 6.9 | **25.6** |

**Quality improvement:** approximately **54%**

## AutoScientist Evaluation

The adapted dataset was used for model training and evaluation.

| Evaluation | Base | Adapted |
|---|---:|---:|
| Overall Win Rate | 10% | **90%** |
| HR Win Rate | 27% | **73%** |

The results show a substantial improvement in HR-focused performance after dataset adaptation.

## Example

```text
Question:
What strategies can organizations use to create personalized
skill development plans?

Answer:
Organizations can assess an employee's current skills, identify
career goals, recommend relevant training and mentoring, establish
measurable milestones, and review progress regularly.
credit: "Adaptive Data by Adaption Labs"


A LORA adapter for `meta-llama/Llama-3.3-70B-Instruct-Reference`. This model was trained with SFT using [Adaption](https://adaptionlabs.ai)'s AutoScientist on the hr_management_qa dataset.

![Training metrics](training-metrics.png)

### AutoScientist Config

```json
{
  "job_id": "3396a5fc-bbc4-4fd0-a331-c447b80d9660",
  "training_experiment_id": "ba32d529-700d-4685-9b4f-c6318459deac",
  "original_model_name": "meta-llama/Llama-3.3-70B-Instruct-Reference",
  "trained_model_name": "adaption_hr_management_qa",
  "training_method": "sft",
  "training_type": "lora",
  "data_format": "chat",
  "hyperparams": {
    "lora": "true",
    "lora_r": 64,
    "n_evals": 5,
    "n_epochs": 3,
    "batch_size": "max",
    "lora_alpha": 128,
    "lora_dropout": 0,
    "min_lr_ratio": 0.1,
    "warmup_ratio": 0.05,
    "weight_decay": 0.02,
    "learning_rate": 0.0001,
    "max_grad_norm": 1,
    "base_model_size": "70B",
    "train_on_inputs": "false",
    "training_method": "sft",
    "lr_scheduler_type": "linear",
    "scheduler_num_cycles": 0.5,
    "lora_trainable_modules": "all-linear"
  }
}
```

## Training Data

The model was trained on 1,254 rows of adapted data with the following domain distribution: hr (91%), legal (2%), corporate-business (2%), career-workplace (1%), governance (1%), personal-finance (1%), technology (1%), academic-education (0%), marketing (0%), data-analysis-visualization (0%), science (0%), personal-growth (0%), architecture-design (0%).

## Model Evaluation

The model was evaluated on an in-distribution held-out test set as well as a broader domain-specific test set to measure generalization.


![Win rates](win-rates.png)

| Domain | Win rate vs. base model |
| --- | --- |
| hr | 73% |

## How to use

```bash
pip install torch transformers peft
```

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

BASE = "meta-llama/Llama-3.3-70B-Instruct-Reference"
ADAPTER = "<this-repo-id>"

device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.float32 if device == "cpu" else torch.bfloat16

base = AutoModelForCausalLM.from_pretrained(BASE, dtype=dtype).to(device)
model = PeftModel.from_pretrained(base, ADAPTER)
# Optional: merge the LoRA weights into the base for faster inference
model = model.merge_and_unload()
model.eval()

tokenizer = AutoTokenizer.from_pretrained(BASE)
messages = [{"role": "user", "content": "Hello!"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(device)

with torch.inference_mode():
    out = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
```