Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -6,31 +6,41 @@ colorTo: blue
|
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
---
|
|
|
|
| 9 |
<div align="center">
|
| 10 |
<h1>🧠 Opus Research</h1>
|
| 11 |
<p><strong>Training AI models from scratch, one parameter at a time.</strong></p>
|
| 12 |
<p><em>"We stand at the right place at the right time."</em></p>
|
| 13 |
</div>
|
| 14 |
-
|
| 15 |
## 👋 About Us
|
|
|
|
| 16 |
We're two teenage AI enthusiasts (ages 15 & 17) passionate about understanding AI from the ground up. Instead of just using pre-trained models, we build them ourselves.
|
|
|
|
| 17 |
## 🚀 Our Models
|
|
|
|
| 18 |
| Model | Parameters | Architecture | Status |
|
| 19 |
|-------|-----------|--------------|--------|
|
| 20 |
| [Opus 1.5](https://huggingface.co/opus-research/opus-1.5) | 0.88B | LLaMA-style | ✅ Released |
|
| 21 |
| Opus 2.0 | 3B+ | LLaMA + Reasoning | 🔜 Coming Soon |
|
|
|
|
| 22 |
## 🔬 Research Focus
|
|
|
|
| 23 |
- **Training from scratch** - No pre-trained weights, 100% original
|
| 24 |
- **Chain-of-thought reasoning** - Teaching models to think before answering
|
| 25 |
- **Efficient architectures** - Sub-3B models that run on consumer GPUs
|
|
|
|
| 26 |
## 📊 Opus 1.5 Highlights
|
|
|
|
| 27 |
- **0.88 billion parameters**
|
| 28 |
- **~2 GB VRAM** for inference
|
| 29 |
- **42 hours** training on 2x RTX 4090
|
| 30 |
- **LLaMA architecture** with RoPE, SwiGLU, GQA, FlashAttention-2
|
|
|
|
| 31 |
## 🔗 Links
|
|
|
|
| 32 |
- 📦 [Opus 1.5 on HuggingFace](https://huggingface.co/opus-research/opus-1.5)
|
| 33 |
-
|
| 34 |
<div align="center">
|
| 35 |
<p>Made with ❤️ and way too much GPU power</p>
|
| 36 |
</div>
|
|
|
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
+
|
| 10 |
<div align="center">
|
| 11 |
<h1>🧠 Opus Research</h1>
|
| 12 |
<p><strong>Training AI models from scratch, one parameter at a time.</strong></p>
|
| 13 |
<p><em>"We stand at the right place at the right time."</em></p>
|
| 14 |
</div>
|
| 15 |
+
|
| 16 |
## 👋 About Us
|
| 17 |
+
|
| 18 |
We're two teenage AI enthusiasts (ages 15 & 17) passionate about understanding AI from the ground up. Instead of just using pre-trained models, we build them ourselves.
|
| 19 |
+
|
| 20 |
## 🚀 Our Models
|
| 21 |
+
|
| 22 |
| Model | Parameters | Architecture | Status |
|
| 23 |
|-------|-----------|--------------|--------|
|
| 24 |
| [Opus 1.5](https://huggingface.co/opus-research/opus-1.5) | 0.88B | LLaMA-style | ✅ Released |
|
| 25 |
| Opus 2.0 | 3B+ | LLaMA + Reasoning | 🔜 Coming Soon |
|
| 26 |
+
|
| 27 |
## 🔬 Research Focus
|
| 28 |
+
|
| 29 |
- **Training from scratch** - No pre-trained weights, 100% original
|
| 30 |
- **Chain-of-thought reasoning** - Teaching models to think before answering
|
| 31 |
- **Efficient architectures** - Sub-3B models that run on consumer GPUs
|
| 32 |
+
|
| 33 |
## 📊 Opus 1.5 Highlights
|
| 34 |
+
|
| 35 |
- **0.88 billion parameters**
|
| 36 |
- **~2 GB VRAM** for inference
|
| 37 |
- **42 hours** training on 2x RTX 4090
|
| 38 |
- **LLaMA architecture** with RoPE, SwiGLU, GQA, FlashAttention-2
|
| 39 |
+
|
| 40 |
## 🔗 Links
|
| 41 |
+
|
| 42 |
- 📦 [Opus 1.5 on HuggingFace](https://huggingface.co/opus-research/opus-1.5)
|
| 43 |
+
|
| 44 |
<div align="center">
|
| 45 |
<p>Made with ❤️ and way too much GPU power</p>
|
| 46 |
</div>
|