Lite Mkd commited on
Update README.md
Browse files
README.md
CHANGED
|
@@ -7,4 +7,61 @@ sdk: static
|
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
+
<div align="center">
|
| 11 |
+
|
| 12 |
+
<h1>EXOFORCE AI</h1>
|
| 13 |
+
|
| 14 |
+
<h3>Building efficient intelligence without datacenter-scale hardware.</h3>
|
| 15 |
+
|
| 16 |
+
<p>
|
| 17 |
+
Language Models · Training · Inference · Agents · Memory · Reinforcement Learning
|
| 18 |
+
</p>
|
| 19 |
+
|
| 20 |
+
<p>
|
| 21 |
+
<a href="https://ai.exoforce.org">Website</a>
|
| 22 |
+
</p>
|
| 23 |
+
|
| 24 |
+
</div>
|
| 25 |
+
|
| 26 |
+
---
|
| 27 |
+
|
| 28 |
+
## What we build
|
| 29 |
+
|
| 30 |
+
EXOFORCE AI is the AI research and engineering division of **EXOFORCE**.
|
| 31 |
+
|
| 32 |
+
We explore how far capable AI systems can be pushed under real hardware constraints — from small language models and experimental architectures to training infrastructure, inference systems, agents, and persistent memory.
|
| 33 |
+
|
| 34 |
+
Our work is centered around systems that can actually be **trained, tested, and deployed without requiring a datacenter**.
|
| 35 |
+
|
| 36 |
+
## Research
|
| 37 |
+
|
| 38 |
+
- **Small Language Models** — capability at compact parameter counts
|
| 39 |
+
- **Efficient Training** — getting more from limited compute and VRAM
|
| 40 |
+
- **Inference** — latency, throughput, memory usage, and deployment
|
| 41 |
+
- **Model Architecture** — experimenting beyond default Transformer recipes
|
| 42 |
+
- **Agents** — models interacting with tools, environments, and other systems
|
| 43 |
+
- **Memory** — long-term and external memory beyond the context window
|
| 44 |
+
- **Reinforcement Learning** — learning through interaction and feedback
|
| 45 |
+
- **Multimodality** — extending models beyond text
|
| 46 |
+
|
| 47 |
+
## Models
|
| 48 |
+
|
| 49 |
+
### Antares
|
| 50 |
+
|
| 51 |
+
Our experimental language-model family focused on efficient architectures, constrained-hardware training, and research at small scale.
|
| 52 |
+
|
| 53 |
+
### Zeon
|
| 54 |
+
|
| 55 |
+
A model family used across EXOFORCE AI for language-model and post-training experiments.
|
| 56 |
+
|
| 57 |
+
---
|
| 58 |
+
|
| 59 |
+
<div align="center">
|
| 60 |
+
|
| 61 |
+
<strong>EXOFORCE AI</strong>
|
| 62 |
+
|
| 63 |
+
<br>
|
| 64 |
+
|
| 65 |
+
<em>Build more with less compute.</em>
|
| 66 |
+
|
| 67 |
+
</div>
|