Use Docker images
docker run --gpus all \
--shm-size 32g \
-p 30000:30000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HF_TOKEN=<secret>" \
--ipc=host \
lmsysorg/sglang:latest \
python3 -m sglang.launch_server \
--model-path "Astrixnet/cilo" \
--host 0.0.0.0 \
--port 30000# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "Astrixnet/cilo",
"messages": [
{
"role": "user",
"content": "What is the capital of France?"
}
]
}'Cilo
A multilingual conversational AI assistant for Indian languages
Developed by Provizoraq Labs — Project Astrix
Overview
Cilo is a 24B-parameter instruction-tuned assistant optimized for natural, helpful conversation across English and major Indian languages. It is designed for production assistant workloads where responsiveness, multilingual fluency, and a consistent assistant persona matter.
Language Support
Cilo supports English and 10 Indic languages:
| English | Hindi | Bengali |
| Gujarati | Kannada | Malayalam |
| Marathi | Odia | Punjabi |
| Tamil | Telugu |
Conversational quality is strongest in English and Hindi; other supported languages are inherited from the base model's broad Indic capabilities.
Highlights
- Multilingual — fluent responses across English and major Indian languages, including code-switching (e.g. Hinglish).
- Instruction-tuned — aligned for clear, task-oriented, conversational responses.
- 24B parameters — strong reasoning and instruction-following at a deployable scale.
- Consistent persona — reliable assistant identity across turns.
Intended Use
Conversational assistants, customer support, education, and general-purpose multilingual text generation.
Out of scope: high-stakes decisions (legal, medical, financial) without human review, and any use prohibited by the license.
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_id = "masterjiii/cilo"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
model_id, torch_dtype=torch.bfloat16, device_map="auto"
)
messages = [
{"role": "system", "content": "You are Cilo, a helpful AI assistant."},
{"role": "user", "content": "Introduce yourself."},
]
inputs = tokenizer.apply_chat_template(
messages, return_tensors="pt", add_generation_prompt=True
).to(model.device)
out = model.generate(inputs, max_new_tokens=256, temperature=0.7, top_p=0.9)
print(tokenizer.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))
Chat Template
Cilo uses a system / user / assistant chat format. Always provide a system message to anchor the assistant persona for best results.
Training
Cilo was instruction-tuned with a curated conversational and identity dataset using parameter-efficient fine-tuning (LoRA), then merged to a standalone model.
Limitations
- May produce inaccurate or outdated information; verify important facts.
- Conversational quality is strongest in English and Hindi.
- Like all LLMs, it can be sensitive to prompt phrasing.
License
Released under the Apache 2.0 license.
Citation
@misc{cilo2025,
title = {Cilo: A Multilingual Conversational Assistant for Indian Languages},
author = {Provizoraq Labs},
year = {2025},
note = {Project Astrix},
url = {https://astrix.network}
}
- Downloads last month
- 114
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "Astrixnet/cilo" \ --host 0.0.0.0 \ --port 30000# Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Astrixnet/cilo", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'