Update README.md
Browse files
README.md
CHANGED
|
@@ -7,4 +7,26 @@ sdk: static
|
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
pinned: false
|
| 8 |
---
|
| 9 |
|
| 10 |
+
# System-L Research
|
| 11 |
+
|
| 12 |
+
LLM pre-training, mid-training, and RL research, active since 2024, led by ShadOwO (previously known as ShadowLilac)
|
| 13 |
+
|
| 14 |
+
Focus: AI anime girl characters and streaming-optimized model development.
|
| 15 |
+
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
## Research areas
|
| 19 |
+
|
| 20 |
+
- Continued pretraining around teaching LLMs VTuber/streamer/VRChat specific knowledge
|
| 21 |
+
- Reward models for character consistency and preference scoring
|
| 22 |
+
- Long-horizon RL around teaching LLMs to be first-person agents in videogames
|
| 23 |
+
- High quality low latency voice generation
|
| 24 |
+
|
| 25 |
+
---
|
| 26 |
+
## Compute resources
|
| 27 |
+
The lab has access to the following compute resources. Additional compute is acquired via Lambda Cloud
|
| 28 |
+
- Nvidia GB10 (2x)
|
| 29 |
+
- i7-13700F, RTX 3090 (2x), 128gb RAM, 4TB
|
| 30 |
+
- 8TB Disk
|
| 31 |
+
|
| 32 |
+
> Models, datasets, and research artifacts published here are part of ongoing work in character-driven LLM development.
|