File size: 3,181 Bytes
dd1064b 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd dd1064b 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c 5c839dd 7eef48c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 | ---
base_model: Qwen/Qwen2.5-3B-Instruct
library_name: peft
pipeline_tag: text-generation
tags:
- base_model:adapter:Qwen/Qwen2.5-3B-Instruct
- lora
- sft
- transformers
- trl
- german
- english
- aether
- gguf
license: apache-2.0
language:
- de
- en
---
# Aether 2.5
Aether 2.5 represents a major milestone in the Aether model series, built on top of the **Qwen2.5-3B-Instruct** base architecture. Trained with SFT (Supervised Fine-Tuning) via Hugging Face TRL and PEFT (LoRA) on a custom 3 GB dataset using local NVIDIA RTX GPU acceleration, Aether 2.5 offers significantly higher intelligence, broader contextual understanding, and superior multilingual responses in German and English.
> 🚀 **Looking for GGUF versions?**
> If you want to run Aether 2.5 locally via **LM Studio**, **Ollama**, or **llama.cpp**, check out the pre-quantized GGUF repository:
> 👉 **[Maxilicious20/Aether-2.5-GGUF](https://huggingface.co/Maxilicious20/Aether-2.5-GGUF)**
## 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/Qwen2.5-3B-Instruct
## Uses
### Direct Use
Aether 2.5 is designed for high-capability conversational AI, complex instruction following, creative text generation, and technical reasoning. Thanks to its LoRA adapter implementation, it delivers flagship 3B-class performance while remaining light enough to run efficiently on local hardware.
### Quantized & GGUF Models
For standalone, CPU/GPU local execution without Python/Transformers dependencies, use the quantized GGUF binaries:
* 📦 **GGUF Repository:** [Maxilicious20/Aether-2.5-GGUF](https://huggingface.co/Maxilicious20/Aether-2.5-GGUF)
* **Available Quantizations:**
* `aether_2_5_fp16.gguf` (Uncompressed / Full Precision)
* `Aether-2.5-3B-Q8_0.gguf` (High Quality / 8-bit)
* `Aether-2.5-3B-Q4_K_M.gguf` (Recommended / Balanced Performance & VRAM)
### How to Get Started with the Model
#### Python (Transformers & PEFT)
Use the following Python code to load Aether 2.5 with `transformers` and `peft`:
```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_model_id = "Qwen/Qwen2.5-3B-Instruct"
adapter_id = "Maxilicious20/Aether-2.5"
# 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 2.5 LoRA Adapter
model = PeftModel.from_pretrained(base_model, adapter_id)
# Example Prompt
messages = [
{"role": "system", "content": "You are Aether 2.5, an advanced AI assistant."},
{"role": "user", "content": "Hello! What improvements do you bring as a 3B model?"}
]
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)) |