Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
|
@@ -4,7 +4,39 @@ emoji: 🏆
|
|
| 4 |
colorFrom: yellow
|
| 5 |
colorTo: yellow
|
| 6 |
sdk: static
|
| 7 |
-
pinned:
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
colorFrom: yellow
|
| 5 |
colorTo: yellow
|
| 6 |
sdk: static
|
| 7 |
+
pinned: true
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# CompactAI
|
| 11 |
+
|
| 12 |
+
We teach tiny neural networks to think. Sometimes they surprise us. Sometimes they just output zeros.
|
| 13 |
+
|
| 14 |
+
## What We Actually Do
|
| 15 |
+
|
| 16 |
+
We train small AI models from absolute scratch. Every weight begins at nothing. Every parameter earns its place through stubborn repetition and repeated failure. We design architectures that fit on a laptop instead of requiring a small moon for cooling. Our creations handle specific tasks without demanding a power plant.
|
| 17 |
+
|
| 18 |
+
## How We Build Them
|
| 19 |
+
|
| 20 |
+
We start with completely blank networks that know nothing about the world. We feed them examples. They guess wrong. We sigh and adjust the math. We repeat the cycle until the loss function stops mocking us. Our training rigs sound like struggling ceiling fans. We optimize every single operation because our budget barely covers decent coffee. The result is a lean system that actually runs on consumer hardware.
|
| 21 |
+
|
| 22 |
+
initialize_weights_to_zero
|
| 23 |
+
for epoch in range our_patience
|
| 24 |
+
train_on_data
|
| 25 |
+
watch_loss_curve_cry
|
| 26 |
+
adjust_learning_rate
|
| 27 |
+
hope_for_the_best
|
| 28 |
+
|
| 29 |
+
## Why We Keep Them Small
|
| 30 |
+
|
| 31 |
+
Massive models consume memory and pretend to understand human emotion. We build modest alternatives. They deploy on everyday machines. They run without requiring a cloud architecture degree. They occasionally mix up a decimal place, but they do so with remarkable efficiency. We believe intelligence works best without a huge data center. We chase practical utility over computational excess.
|
| 32 |
+
|
| 33 |
+
* They load instantly on mobile devices
|
| 34 |
+
* They cost pennies to run at scale
|
| 35 |
+
* They never ask for your life story
|
| 36 |
+
* They respect your limited RAM
|
| 37 |
+
|
| 38 |
+
## Getting Started
|
| 39 |
+
|
| 40 |
+
Read the documentation if we have written any. Download a model if we have released one. Run it on your machine. Watch it do something small, but do it efficiently. If nothing works, blame the random seed.
|
| 41 |
+
|
| 42 |
+
We are CompactAI. We make small things work. We apologize in advance when they do not. jk.
|