131 lines
3.4 KiB
Markdown
131 lines
3.4 KiB
Markdown
|
---
|
||
|
|
base_model: unsloth/Qwen2.5-7B-Instruct # Уточни, если база другая
|
||
|
|
tags:
|
||
|
|
- text-generation-inference
|
||
|
|
- transformers
|
||
|
|
- unsloth
|
||
|
|
- qwen2
|
||
|
|
- reasoning
|
||
|
|
- science
|
||
|
|
- logic
|
||
|
|
- trl
|
||
|
|
- sft
|
||
|
|
license: apache-2.0
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
- ru
|
||
|
|
- zh
|
||
|
|
model-index:
|
||
|
|
- name: Noir-Ultra
|
||
|
|
results:
|
||
|
|
- task:
|
||
|
|
type: text-generation
|
||
|
|
name: STEM (Science)
|
||
|
|
dataset:
|
||
|
|
name: SciQ
|
||
|
|
type: sciq
|
||
|
|
metrics:
|
||
|
|
- name: accuracy
|
||
|
|
type: exact_match
|
||
|
|
value: 91.0
|
||
|
|
- task:
|
||
|
|
type: text-generation
|
||
|
|
name: Mathematics
|
||
|
|
dataset:
|
||
|
|
name: GSM8K
|
||
|
|
type: gsm8k
|
||
|
|
metrics:
|
||
|
|
- name: accuracy
|
||
|
|
type: exact_match
|
||
|
|
value: 84.0
|
||
|
|
- task:
|
||
|
|
type: text-generation
|
||
|
|
name: Logic
|
||
|
|
dataset:
|
||
|
|
name: ARC Challenge
|
||
|
|
type: ai2_arc
|
||
|
|
metrics:
|
||
|
|
- name: accuracy
|
||
|
|
type: exact_match
|
||
|
|
value: 86.0
|
||
|
|
---
|
||
|
|
|
||
|
|
# 🌌 Noir-Ultra (7B)
|
||
|
|
|
||
|
|
<div align="center">
|
||
|
|
|
||
|
|
[Noir Family](https://huggingface.co/collections/muverqqw/noir) | [GGUF Versions](https://huggingface.co/muverqqw/Noir-GGUF) | [Technical Report](#-noir-ultra-evaluation-dashboard)
|
||
|
|
|
||
|
|
</div>
|
||
|
|
|
||
|
|
**Noir-Ultra** is the elite 7-billion parameter model of the Noir series. It represents a breakthrough in training efficiency: where previous 7B iterations required 6 epochs to reach stability, **Noir-Ultra achieved superior results in just a single epoch**.
|
||
|
|
|
||
|
|
This model is a "compact titan," delivering scientific reasoning and mathematical accuracy that rival much larger architectures.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 🚀 The Ultra Advantage
|
||
|
|
|
||
|
|
* 🧬 **Unrivaled STEM:** With a **91.0% score on SciQ**, it is a specialized tool for scientific inquiry.
|
||
|
|
* 📐 **Mathematical Precision:** Scoring **84.0% on GSM8K**, it handles complex chains of thought with ease.
|
||
|
|
* 🧠 **Logical Depth:** An ARC-Challenge score of **86.0%** places it at the top of its weight class for reasoning tasks.
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 📊 Evaluation Dashboard (Noir Ultra Report)
|
||
|
|
|
||
|
|
Based on the latest evaluation, Noir-Ultra shows an exceptionally strong profile in technical domains:
|
||
|
|
|
||
|
|
| Domain | Benchmark | Result (%) | Status |
|
||
|
|
| :--- | :--- | :---: | :--- |
|
||
|
|
| **STEM** | SciQ | **91.0%** | 🏆 Master |
|
||
|
|
| **Logic** | ARC-C | **86.0%** | 🔥 Elite |
|
||
|
|
| **Math** | GSM8K | **84.0%** | ✅ Advanced |
|
||
|
|
| **Medicine** | MedQA | **65.0%** | 🩺 Competent |
|
||
|
|
| **Physics** | MMLU-Physics| **70.0%** | 🧪 Specialist |
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 📦 Noir Model Family Matrix
|
||
|
|
|
||
|
|
| Model | Parameters | Role | Key Strength |
|
||
|
|
| :--- | :--- | :--- | :--- |
|
||
|
|
| **Noir-Lightning** | 0.5B | The Pocket Assistant | Ultra-fast, runs on anything |
|
||
|
|
| **Noir-Mini** | 1.5B | The Balanced Thinker | High speed with solid grammar |
|
||
|
|
| **Noir-Standard** | 3B | The Versatile Workhorse | 65% GSM8K, perfect for 8GB VRAM |
|
||
|
|
| **Noir-Ultra** | **7B** | **The Reasoning Master** | **91% SciQ & 84% Math** |
|
||
|
|
| **Noir-Starlight** | 14B | The Galactic Intelligence | Deep logic & Expert-level STEM |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## 🛠 Quick Start (Transformers)
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
import torch
|
||
|
|
|
||
|
|
model_name = "muverqqw/Noir-Ultra"
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
model_name,
|
||
|
|
load_in_4bit=True, # Recommended for 8GB VRAM
|
||
|
|
device_map="auto"
|
||
|
|
)
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
### 👤 Credits
|
||
|
|
* **Developer:** IceL1ghtning
|
||
|
|
|
||
|
|
* **Architecture:** Qwen 2.5 (7B)
|
||
|
|
|
||
|
|
* **Release:** 2026
|
||
|
|
|
||
|
|
* **License:** Apache 2.0
|
||
|
|
|
||
|
|
<div align="center">
|
||
|
|
<sub>Efficiency meets intelligence. Built with passion for the open-source community.</sub>
|
||
|
|
</div>
|