---
base_model: Qwen/Qwen2.5-1.5B-Instruct
library_name: peft
pipeline_tag: text-generation
tags:
- base_model:adapter:Qwen/Qwen2.5-1.5B-Instruct
- lora
- sft
- transformers
- trl
- german
- english
- aether
license: apache-2.0
language:
- de
- en
---

# Aether 2.2

Aether 2.2 is a lightweight, fine-tuned language model based on **Qwen2.5-1.5B-Instruct**. It was optimized using PEFT (LoRA) to deliver improved response quality, instruction following, and conversational fluency in both German and English while maintaining minimal VRAM usage.

## Model Details

### Model Description

- **Developed by:** Maxilicious20
- **Model type:** Causal Language Model (LoRA Adapter)
- **Language(s) (NLP):** German, English
- **License:** Apache-2.0
- **Finetuned from model:** Qwen/Qwen2.5-1.5B-Instruct

## Uses

### Direct Use

This model is designed as an intelligent assistant for text generation, conversational chat, and general reasoning tasks. Thanks to its lightweight LoRA adapter format, it can be run locally with minimal VRAM requirements.

### How to Get Started with the Model

Use the following Python code with `transformers` and `peft` to load Aether 2.2 directly:

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

base_model_id = "Qwen/Qwen2.5-1.5B-Instruct"
adapter_id = "Maxilicious20/Aether-2.2"

# Load Tokenizer and Base Model
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Load Aether 2.2 LoRA Adapter
model = PeftModel.from_pretrained(base_model, adapter_id)

# Example Prompt
messages = [
    {"role": "system", "content": "You are Aether, a helpful AI assistant."},
    {"role": "user", "content": "Hello! Who are you and what can you do?"}
]

prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))