Spaces:
Running
Running
Update README.md
Browse files
README.md
CHANGED
@@ -17,43 +17,12 @@ We believe that meaningful storytelling goes beyond just words β it's about un
|
|
17 |
## Latest Models
|
18 |
|
19 |
### π Cakrawala-70B
|
20 |
-
Our flagship model, built on Llama-3.1-70B-Instruct, specifically optimized for emotional intelligence in storytelling. Trained on 8 x H100 NVL GPUs. It's specifically optimised for generating rich roleplaying conversations and character interactions. The model has been trained to excel at producing detailed, contextually appropriate character dialogues with rich descriptions of physical actions, expressions, and emotional states while maintaining consistent character voices and perspectives
|
21 |
-
|
22 |
-
#### Training Specifications:
|
23 |
-
- Specialized dataset of 5,867 emotionally rich conversations
|
24 |
-
- Minimum 12-13 turns per conversation for deep emotional development
|
25 |
-
- Fine-tuned using QLoRA over 3 epochs
|
26 |
-
- Focus on emotional consistency and character growth
|
27 |
-
|
28 |
-
#### Technical Parameters:
|
29 |
-
- Gradient Accumulation Steps: 16
|
30 |
-
- Micro Batch Size: 4
|
31 |
-
- Learning Rate: 0.0003
|
32 |
-
- Optimizer: AdamW
|
33 |
-
- Mixed Precision: BF16 & FP16 with TF32 support
|
34 |
|
35 |
### π Cakrawala-8B
|
36 |
Our compact model built on Llama-3.1-8B-Instruct, delivering high quality roleplay in a lighter model. Trained on 2 x H100 SXM GPUs.
|
37 |
|
38 |
-
## What Sets Us Apart
|
39 |
-
|
40 |
-
### Emotional Depth Training
|
41 |
-
Our models are specifically trained to understand:
|
42 |
-
- Complex emotional states and transitions
|
43 |
-
- Non-verbal emotional cues
|
44 |
-
- Psychological motivation and character growth
|
45 |
-
- Consistent emotional growth arcs
|
46 |
-
- Memory of emotional experiences
|
47 |
-
- Authentic personality expression
|
48 |
-
- Behavioral consistency with emotional state
|
49 |
-
|
50 |
-
## License & Credits
|
51 |
-
- Licensed under MIT
|
52 |
-
- Based on meta-llama/Llama-3.1 models
|
53 |
-
- *Created with empathy, driven by understanding*
|
54 |
-
|
55 |
-
---
|
56 |
|
57 |
-
*Join us in revolutionizing AI
|
58 |
|
59 |
-
For more information, collaboration opportunities, or to contribute to our mission, please reach out to our
|
|
|
17 |
## Latest Models
|
18 |
|
19 |
### π Cakrawala-70B
|
20 |
+
Our flagship model, built on Llama-3.1-70B-Instruct, specifically optimized for emotional intelligence in storytelling. Trained on 8 x H100 NVL GPUs. It's specifically optimised for generating rich roleplaying conversations and character interactions. The model has been trained to excel at producing detailed, contextually appropriate character dialogues with rich descriptions of physical actions, expressions, and emotional states while maintaining consistent character voices and perspectives.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
|
22 |
### π Cakrawala-8B
|
23 |
Our compact model built on Llama-3.1-8B-Instruct, delivering high quality roleplay in a lighter model. Trained on 2 x H100 SXM GPUs.
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
+
*Join us in revolutionizing AI roleplay through emotional intelligence and deeper character understanding.*
|
27 |
|
28 |
+
For more information, collaboration opportunities, or to contribute to our mission, please reach out to our Discord.
|