File size: 1,925 Bytes
220c7b1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 | ---
license: apache-2.0
language:
- en
- ru
tags:
- transformers
- safetensors
- gemma3
- image-text-to-text
- creative
- roleplay
- conversational
- text-generation-inference
- endpoints_compatible
base_model:
- OddTheGreat/Mars_27B_V.1
model_name: Majuli 3.1
pipeline_tag: image-text-to-text
---
# Majuli 3.1
**By Tripplet AI** (Tripplet Artificial General Intelligence Research Institute)
Majuli 3.1 is a powerful 27B parameter multimodal language model built on the Gemma 3 architecture, optimized for creative writing, roleplay, and general-purpose instruction following.
## Model Details
- **Parameters:** 28.8B
- **Architecture:** Gemma 3 (Gemma3ForConditionalGeneration)
- **Context Length:** 131,072 tokens
- **Hidden Size:** 5376
- **Layers:** 62
- **Attention Heads:** 32 (16 KV heads)
- **Vision Encoder:** SigLIP (896px, 27 layers)
- **Languages:** English, Russian
- **Precision:** bfloat16
## Key Features
- Long context support up to 128K tokens
- Multimodal capabilities (image + text)
- Hybrid attention with sliding window (1024) and full attention layers
- Optimized for creative and roleplay tasks
## Usage
```python
from transformers import AutoProcessor, AutoModelForImageTextToText
model = AutoModelForImageTextToText.from_pretrained("tripplet-research/majuli-3.1")
processor = AutoProcessor.from_pretrained("tripplet-research/majuli-3.1")
messages = [
{"role": "user", "content": "Hello, tell me about yourself."}
]
inputs = processor.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
output = model.generate(**inputs, max_new_tokens=512)
print(processor.decode(output[0], skip_special_tokens=True))
```
## About Tripplet AI
Tripplet Artificial General Intelligence Research Institute is dedicated to advancing the frontiers of artificial general intelligence through open research and model development.
## License
Apache 2.0
|