Update README.md
Browse files
README.md
CHANGED
|
@@ -1,10 +1,13 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
|
|
|
| 4 |
# JEPA-Style LLM Prototypes
|
| 5 |
|
| 6 |
Making decoder-only transformers predict state consequences instead of tokens.
|
| 7 |
|
|
|
|
|
|
|
| 8 |
## What's This?
|
| 9 |
|
| 10 |
Three approaches to convert a standard LLM into a world model that predicts "what happens next" given a state and action — like JEPA but for language models.
|
|
@@ -20,7 +23,7 @@ Three approaches to convert a standard LLM into a world model that predicts "wha
|
|
| 20 |
## Quick Start
|
| 21 |
|
| 22 |
1. Open any notebook in [Google Colab](https://colab.research.google.com/)
|
| 23 |
-
2. Set runtime to **GPU** (Runtime → Change runtime type →
|
| 24 |
3. Run all cells
|
| 25 |
4. Watch the model learn to predict state transitions
|
| 26 |
|
|
@@ -80,4 +83,6 @@ All dependencies install automatically in the notebooks.
|
|
| 80 |
|
| 81 |
---
|
| 82 |
|
| 83 |
-
*Experimental code — have fun breaking it.*
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
| 4 |
+
|
| 5 |
# JEPA-Style LLM Prototypes
|
| 6 |
|
| 7 |
Making decoder-only transformers predict state consequences instead of tokens.
|
| 8 |
|
| 9 |
+
🔗 **[View on Hugging Face](https://huggingface.co/wassemgtk/jepa_llm_prototypes)**
|
| 10 |
+
|
| 11 |
## What's This?
|
| 12 |
|
| 13 |
Three approaches to convert a standard LLM into a world model that predicts "what happens next" given a state and action — like JEPA but for language models.
|
|
|
|
| 23 |
## Quick Start
|
| 24 |
|
| 25 |
1. Open any notebook in [Google Colab](https://colab.research.google.com/)
|
| 26 |
+
2. Set runtime to **GPU** (Runtime → Change runtime type → T4)
|
| 27 |
3. Run all cells
|
| 28 |
4. Watch the model learn to predict state transitions
|
| 29 |
|
|
|
|
| 83 |
|
| 84 |
---
|
| 85 |
|
| 86 |
+
*Experimental code — have fun breaking it.*
|
| 87 |
+
|
| 88 |
+
**Coauthors:** Writer Agent & OpenCode
|