Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -10,5 +10,87 @@ pinned: false
|
|
| 10 |
license: mit
|
| 11 |
short_description: Chat with Darwin-Qwen3-4B evolutionary merged model
|
| 12 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
license: mit
|
| 11 |
short_description: Chat with Darwin-Qwen3-4B evolutionary merged model
|
| 12 |
---
|
| 13 |
+
<div align="center">
|
| 14 |
+
<span style="font-family: default; font-size: 1.5em;">Darwin-Qwen3-4B</span>
|
| 15 |
+
<div>
|
| 16 |
+
evolutionary algorithm 'Darwin A2AP' 🤔
|
| 17 |
+
</div>
|
| 18 |
+
</div>
|
| 19 |
+
<br>
|
| 20 |
+
<div align="center" style="line-height: 1;">
|
| 21 |
+
<a href=" https://discord.gg/openfreeai" style="margin: 2px;">
|
| 22 |
+
<img alt="OpenFree AI Discord Server" src="https://img.shields.io/badge/Discord-000000?style=for-the-badge&logo=discord&logoColor=000&logoColor=white" style="display: inline-block; vertical-align: middle;"/>
|
| 23 |
+
</a>
|
| 24 |
+
<a href="https://huggingface.co/VIDraft" style="margin: 2px;">
|
| 25 |
+
<img alt="HF Page" src="https://img.shields.io/badge/VIDraft-fcd022?style=for-the-badge&logo=huggingface&logoColor=000&labelColor" style="display: inline-block; vertical-align: middle;"/>
|
| 26 |
+
</a>
|
| 27 |
+
</div>
|
| 28 |
|
| 29 |
+
# openfree/Darwin-Qwen3-4B
|
| 30 |
+
This model is automatically merged using evolutionary algorithm 'Darwin A2AP' v3.2
|
| 31 |
+
|
| 32 |
+
# Overview
|
| 33 |
+
This study introduces a new paradigm of AI model fusion. Traditional "model merging" techniques have been restricted to models of the same family (e.g., transformer-based LLMs). We transcend this limitation by proposing a method to directly collide and fuse the core representational structures (DNA) of entirely different species — such as transformers and diffusion models. This approach acts as an "AI particle accelerator," colliding fundamentally distinct elements of intelligence to uncover new possibilities.
|
| 34 |
+
The paper and source code (to be released on GitHub and Hugging Face) are currently under preparation and will be made publicly available soon. They will be released in a reproducible and extensible form for anyone to explore.
|
| 35 |
+
|
| 36 |
+
## Contribution
|
| 37 |
+
Breaking the Species Barrier
|
| 38 |
+
Fusion of fundamentally different models such as transformers and diffusion architectures.
|
| 39 |
+
Realization of cross-species model merging once deemed impossible.
|
| 40 |
+
|
| 41 |
+
## AI Embryo Creation
|
| 42 |
+
Formation of an initial “AI embryo” based on fused DNA.
|
| 43 |
+
The embryo is not confined to a single domain or function but serves as the foundation for multi-capability intelligence.
|
| 44 |
+
|
| 45 |
+
## Virtual Evolutionary Environment
|
| 46 |
+
AI embryos are placed into a simulated environment spanning thousands of generations.
|
| 47 |
+
Through survival and adaptation, natural selection drives evolution beyond the limitations of parent models, producing new offspring models.
|
| 48 |
+
|
| 49 |
+
## Merge Information
|
| 50 |
+
Father Model 1: Qwen/Qwen3-4B-Instruct-2507
|
| 51 |
+
Mother Model 2: Qwen/Qwen3-4B-Thinking-2507
|
| 52 |
+
Validation Task Accuracy: 88.56%
|
| 53 |
+
Note: The above accuracy is a proxy metric used for merge ratio optimization.
|
| 54 |
+
Algorithm Version: Darwin A2AP Enhanced v3.2
|
| 55 |
+
|
| 56 |
+
## ⚠️ Notice
|
| 57 |
+
The actual language generation performance of this model requires separate evaluation.
|
| 58 |
+
The validation score above is not an LLM benchmark score.
|
| 59 |
+
|
| 60 |
+
## ⚠️ Benchmarking Test Results
|
| 61 |
+
<p align="center"> <img src="BenchmarkResult.png" alt="Darwin-Qwen3-4B BenchMark Result" width="600"/> </p>
|
| 62 |
+
|
| 63 |
+
|
| 64 |
+
## Use_Example
|
| 65 |
+
|
| 66 |
+
```python
|
| 67 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 68 |
+
|
| 69 |
+
model = AutoModelForCausalLM.from_pretrained("openfree/Darwin-Qwen3-4B")
|
| 70 |
+
tokenizer = AutoTokenizer.from_pretrained("openfree/Darwin-Qwen3-4B")
|
| 71 |
+
|
| 72 |
+
# 추론 예시
|
| 73 |
+
inputs = tokenizer("Hello, how are you?", return_tensors="pt")
|
| 74 |
+
outputs = model.generate(**inputs)
|
| 75 |
+
```
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
# Strengths & Features
|
| 79 |
+
## Cross-Domain Intelligence
|
| 80 |
+
Example: Legal LLM × Medical LLM → instantly produces a “Forensic LLM.”
|
| 81 |
+
This is not mere knowledge aggregation but the creation of new intelligence at the intersection of domains.
|
| 82 |
+
|
| 83 |
+
## Extreme Efficiency
|
| 84 |
+
Achieves results at roughly 1/10,000 of the time and cost compared to training a new foundation model.
|
| 85 |
+
Accessible via a simple click-based process.
|
| 86 |
+
|
| 87 |
+
## Unified Intelligence
|
| 88 |
+
Escapes confinement to a single domain by organically merging multiple expertises.
|
| 89 |
+
Provides an experimental basis for integrated reasoning and creativity with AGI-like qualities.
|
| 90 |
+
|
| 91 |
+
## Reproducibility & Openness
|
| 92 |
+
Source code and models will be fully released on GitHub and Hugging Face.
|
| 93 |
+
Researchers and developers can freely reproduce, experiment, and expand.
|
| 94 |
+
|
| 95 |
+
# Outlook
|
| 96 |
+
This research opens the door to a new generation of model creation, expressed as “Foundation a + Foundation b = Foundation abXc.” It represents far more than a reduction in training costs, serving as a critical turning point for future studies on the evolution and fusion of AI intelligence.
|