Update README.md

README.md

@@ -1,57 +1,90 @@
 ---
-language:
-- en
-license: apache-2.0
 base_model: Qwen/Qwen2.5-3B-Instruct
+library_name: peft
+pipeline_tag: text-generation
 tags:
-- mono-ai
-- aether
-- text-generation
+- base_model:adapter:Qwen/Qwen2.5-3B-Instruct
 - lora
-- merged
-pipeline_tag: text-generation
+- sft
+- transformers
+- trl
+- german
+- english
+- aether
+- gguf
+license: apache-2.0
+language:
+- de
+- en
 ---
 
-# 🌌 Aether 2.5
+# 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.
 
-**Aether 2.5** is a fine-tuned 3-billion parameter instruction-following language model developed by **Mono AI Studio**. It is based on `Qwen/Qwen2.5-3B-Instruct` and optimized for enhanced reasoning, knowledge retrieval, and standard text benchmarks.
+> 🚀 **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)**
 
-## 📊 Benchmark Results
+## Model Details
 
-Evaluated on a 6-benchmark evaluation suite (300 MC samples per benchmark):
+### Model Description
 
-| Benchmark | Aether 2.3 | Aether 2.5 | Improvement |
-| :--- | :---: | :---: | :---: |
-| **ARC-Challenge** | 45.3% | **48.3%** | +3.0% |
-| **OpenBookQA** | 49.7% | **52.3%** | +2.6% |
-| **MMLU** | 37.3% | **39.7%** | +2.4% |
-| **HellaSwag** | 51.7% | **54.0%** | +2.3% |
-| **Overall Average** | 33.2% | **33.6%** | **+0.4%** |
+- **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
 
-## 🚀 Usage
+## Uses
 
-You can run this model locally using standard Hugging Face Transformers:
+### 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
-from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
 
-model_name = "Maxilicious20/Aether-2.5"
+base_model_id = "Qwen/Qwen2.5-3B-Instruct"
+adapter_id = "Maxilicious20/Aether-2.5"
 
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16,
+# 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 a highly capable AI assistant developed by Mono AI Studio."},
-    {"role": "user", "content": "Explain quantum computing in simple terms."}
+    {"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=512)
+outputs = model.generate(**inputs, max_new_tokens=256)
 print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))