Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,57 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
| 2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
license: apache-2.0
|
| 5 |
+
base_model: Qwen/Qwen2.5-3B-Instruct
|
| 6 |
+
tags:
|
| 7 |
+
- mono-ai
|
| 8 |
+
- aether
|
| 9 |
+
- text-generation
|
| 10 |
+
- lora
|
| 11 |
+
- merged
|
| 12 |
+
pipeline_tag: text-generation
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
# 🌌 Aether 2.5
|
| 16 |
+
|
| 17 |
+
**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.
|
| 18 |
+
|
| 19 |
+
## 📊 Benchmark Results
|
| 20 |
+
|
| 21 |
+
Evaluated on a 6-benchmark evaluation suite (300 MC samples per benchmark):
|
| 22 |
+
|
| 23 |
+
| Benchmark | Aether 2.3 | Aether 2.5 | Improvement |
|
| 24 |
+
| :--- | :---: | :---: | :---: |
|
| 25 |
+
| **ARC-Challenge** | 45.3% | **48.3%** | +3.0% |
|
| 26 |
+
| **OpenBookQA** | 49.7% | **52.3%** | +2.6% |
|
| 27 |
+
| **MMLU** | 37.3% | **39.7%** | +2.4% |
|
| 28 |
+
| **HellaSwag** | 51.7% | **54.0%** | +2.3% |
|
| 29 |
+
| **Overall Average** | 33.2% | **33.6%** | **+0.4%** |
|
| 30 |
+
|
| 31 |
+
## 🚀 Usage
|
| 32 |
+
|
| 33 |
+
You can run this model locally using standard Hugging Face Transformers:
|
| 34 |
+
|
| 35 |
+
```python
|
| 36 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 37 |
+
import torch
|
| 38 |
+
|
| 39 |
+
model_name = "Maxilicious20/Aether-2.5"
|
| 40 |
+
|
| 41 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 42 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 43 |
+
model_name,
|
| 44 |
+
torch_dtype=torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16,
|
| 45 |
+
device_map="auto"
|
| 46 |
+
)
|
| 47 |
+
|
| 48 |
+
messages = [
|
| 49 |
+
{"role": "system", "content": "You are a highly capable AI assistant developed by Mono AI Studio."},
|
| 50 |
+
{"role": "user", "content": "Explain quantum computing in simple terms."}
|
| 51 |
+
]
|
| 52 |
+
|
| 53 |
+
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
| 54 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
| 55 |
+
|
| 56 |
+
outputs = model.generate(**inputs, max_new_tokens=512)
|
| 57 |
+
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))
|