majuli3.1

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

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