96 lines
2.4 KiB
Markdown
96 lines
2.4 KiB
Markdown
|
---
|
||
|
|
base_model: unsloth/qwen2.5-coder-1.5b-instruct
|
||
|
|
tags:
|
||
|
|
- Omnionix
|
||
|
|
- Avara
|
||
|
|
- qwen2
|
||
|
|
- code
|
||
|
|
- math
|
||
|
|
- logic
|
||
|
|
- transformers
|
||
|
|
- unsloth
|
||
|
|
num_parameters: 2000000000
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
|
||
|
|
<br>
|
||
|
|
|
||
|
|
<p align="center">
|
||
|
|
<img src="logo.png" width="450" alt="Avara X1 Mini Logo">
|
||
|
|
</p>
|
||
|
|
|
||
|
|
<br>
|
||
|
|
|
||
|
|
# Avara X1 Mini
|
||
|
|
|
||
|
|
Avara X1 Mini is a lightweight AI model developed by **Omnionix**. Based on the Qwen2.5 architecture, this model is fine-tuned to balance technical reasoning with a grounded and supportive personality.
|
||
|
|
|
||
|
|
**Join the Community:** [Omnionix Discord](https://discord.gg/mGgRgz6g)
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
### Technical Specifications
|
||
|
|
|
||
|
|
| Feature | Details |
|
||
|
|
| :--- | :--- |
|
||
|
|
| **Developer** | Omnionix |
|
||
|
|
| **Architecture** | Qwen2.5-1.5B |
|
||
|
|
| **Format** | ChatML |
|
||
|
|
| **Identity** | Native Omnionix system logic |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
### Training Methodology
|
||
|
|
|
||
|
|
Avara X1 Mini was fine-tuned using the Unsloth library on a high-density dataset blend designed for maximum reasoning performance in a small footprint:
|
||
|
|
|
||
|
|
* **Code:** The Stack (BigCode) for professional-grade programming logic.
|
||
|
|
* **Mathematics:** Focused math/competition datasets for step-by-step problem solving.
|
||
|
|
* **Logic:** Open-Platypus for enhanced deductive reasoning and instruction following.
|
||
|
|
|
||
|
|
We also have the [LoRA adapter ](https://huggingface.co/Omnionix12345/avara-x1-mini-lora) and the Q4_K_M GGUF: huggingface.co/Omnionix12345/avara-x1-mini-Q4_K_M-GGUF
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
### Implementation
|
||
|
|
|
||
|
|
To use Avara locally, the following standard chat script provides a natural back-and-forth dialogue by managing conversation history automatically.
|
||
|
|
|
||
|
|
```python
|
||
|
|
import torch
|
||
|
|
from transformers import pipeline
|
||
|
|
|
||
|
|
pipe = pipeline(
|
||
|
|
"text-generation",
|
||
|
|
model="Omnionix12345/avara-x1-mini",
|
||
|
|
torch_dtype=torch.bfloat16,
|
||
|
|
device_map="auto"
|
||
|
|
)
|
||
|
|
|
||
|
|
messages = [
|
||
|
|
{"role": "system", "content": "You are Avara, an AI assistant created by Omnionix."}
|
||
|
|
]
|
||
|
|
|
||
|
|
print("\n--- Avara X1 Mini is Online ---")
|
||
|
|
|
||
|
|
while True:
|
||
|
|
user_input = input("You: ")
|
||
|
|
if user_input.lower() in ["exit", "quit"]:
|
||
|
|
break
|
||
|
|
|
||
|
|
messages.append({"role": "user", "content": user_input})
|
||
|
|
|
||
|
|
outputs = pipe(
|
||
|
|
messages,
|
||
|
|
max_new_tokens=512,
|
||
|
|
do_sample=True,
|
||
|
|
temperature=0.7
|
||
|
|
)
|
||
|
|
|
||
|
|
assistant_response = outputs[0]["generated_text"][-1]["content"]
|
||
|
|
print(f"\nAvara: {assistant_response}\n")
|
||
|
|
|
||
|
|
messages.append({"role": "assistant", "content": assistant_response})
|