QatrixInfotech/Cella-Omin1.1

---

Cella Omin 1.1

Cella Omin 1.1 is a multilingual AI assistant developed and fine-tuned by Qatrix Infotech.

The model is designed for general conversation, programming assistance, logical reasoning, writing, text summarization, design discussions, and multilingual interaction.

Model Details

Specification	Details
Model Name	Cella Omin 1.1
Developer	Qatrix Infotech
Parameters	Approximately 4 billion
Format	GGUF
Quantization	Q4_K_M
Primary File	Cella-Omin-1.1-Q4_K_M.gguf
Task	Text generation
License	Apache 2.0

Core Identity

Cella Omin 1.1 includes a baked-in identity established through supervised fine-tuning.

The identity adapter was merged directly into the model weights before GGUF conversion. A separate system prompt is therefore not normally required for basic identity questions.

When asked about its identity, the model is designed to respond:

I am Cella Omin 1.1, a multilingual AI assistant developed by Qatrix Infotech.

Identity consistency may vary depending on the prompt, chat template, inference software, and generation settings.

Capabilities

Cella Omin 1.1 is designed to assist with:

• General conversation
• Multilingual conversation
• Translation assistance
• Programming assistance
• Basic code debugging
• Writing and rewriting
• Text summarization
• Logical reasoning
• Mathematical reasoning
• UI and UX discussions
• Design-related discussions
• Educational tasks
• Local AI experiments

Supported Languages

The model has been tested for basic identity responses and general interaction in the following languages:

• English
• Malayalam
• Tamil
• Kannada
• Telugu
• Urdu
• Punjabi
• Bengali
• Arabic

Performance may vary depending on the language, dialect, prompt structure, subject, and task complexity.

The model may understand additional languages, but those languages have not been tested to the same extent.

Available Files

Cella-Omin-1.1-Q4_K_M.gguf

The primary quantized model file.

Use this file for:

• Chat
• Writing
• Coding assistance
• Reasoning
• Summarization
• Translation
• Multilingual text generation

Recommended Generation Parameters

Parameter	Recommended Value
Temperature	1.0
Top K	64
Top P	0.95
Context Size	8192
Repetition Penalty	1.0

These values are starting recommendations. Different applications may require different settings.

Lower temperatures may produce more predictable responses, while higher temperatures may produce more varied responses.

Running with llama.cpp

Install or build a recent version of llama.cpp.

Run the model using:

llama-cli \
  --model Cella-Omin-1.1-Q4_K_M.gguf \
  --ctx-size 8192 \
  --temp 1.0 \
  --top-k 64 \
  --top-p 0.95 \
  --repeat-penalty 1.0 \
  --conversation

Depending on your version of llama.cpp, command names and supported options may differ.

Running with Ollama

Create a file named Modelfile in the same directory as the GGUF file.

Add the following content:

FROM ./Cella-Omin-1.1-Q4_K_M.gguf

PARAMETER temperature 1.0
PARAMETER top_k 64
PARAMETER top_p 0.95
PARAMETER num_ctx 8192
PARAMETER repeat_penalty 1.0

Create the Ollama model:

ollama create cella-omin-1.1 -f Modelfile

Run the model:

ollama run cella-omin-1.1

Example Prompts

Identity

Who are you?

Writing

Rewrite this paragraph in a more professional tone.

Programming

Explain this Python error and suggest a fix.

Summarization

Summarize the following text in five bullet points.

Translation

Translate the following English paragraph into Malayalam.

Reasoning

Explain your answer clearly and solve the problem step by step.

Intended Uses

Cella Omin 1.1 is intended for local and offline AI workflows, including:

• Personal AI assistants
• Multilingual chat applications
• Educational demonstrations
• Writing assistance
• Programming assistance
• Translation experiments
• Local text-generation applications
• Research prototypes
• AI interface testing
• Prompt-engineering experiments

Out-of-Scope Uses

The model should not be used as the sole source of information for:

• Medical diagnosis or treatment
• Legal decisions
• Financial decisions
• Emergency response
• Safety-critical systems
• High-risk automated decision-making
• Production code without human review

The model should not be used to facilitate unlawful activity, abuse, deception, privacy violations, harassment, or harm.

Training Summary

Cella Omin 1.1 is an early release in the development of the Cella model family.

Training Detail	Description
Framework	Unsloth
Training Method	Supervised Fine-Tuning
Adapter Method	LoRA
Training Precision	4-bit
Hardware	NVIDIA Tesla T4 GPU
Training Focus	Multilingual identity, capability boundaries, and identity-consistency repair
Export Format	GGUF
Quantization	Q4_K_M

The trained LoRA adapter was merged into the model weights before conversion to GGUF.

Dataset Summary

The fine-tuning dataset included examples related to:

• Model identity
• Developer attribution
• Multilingual identity responses
• General assistant behavior
• Capability boundaries
• Identity consistency
• Targeted repair prompts

The dataset focused primarily on improving identity consistency and basic assistant behavior rather than creating a large-scale general-knowledge model.

Limitations

Cella Omin 1.1 is an experimental early release.

Hallucinations

The model may generate incorrect, outdated, misleading, biased, or fabricated information.

Important claims should be independently verified.

Limited Knowledge

The model does not have live internet access unless it is connected to an external search, retrieval, or tool system.

Its internal knowledge may be incomplete or outdated.

Language Variance

Quality may vary across languages.

Performance may decrease with:

• Uncommon dialects
• Mixed-language prompts
• Transliteration
• Complex technical content
• Long conversations
• Ambiguous instructions

Prompt Sensitivity

Responses may change depending on:

• The chat template
• The system prompt
• Prompt wording
• Context length
• Inference software
• Sampling settings

Identity behavior may become inconsistent when the model receives aggressive system instructions, conflicting identity instructions, unusual prompts, or incompatible chat templates.

Code Reliability

Generated code may contain:

• Syntax errors
• Security issues
• Incorrect assumptions
• Outdated APIs
• Missing dependencies
• Logical errors

All generated code should be reviewed and tested before use.

Reasoning Reliability

The model may produce confident explanations containing incorrect reasoning.

Step-by-step answers should not automatically be treated as correct.

Context Limitations

Actual usable context length may depend on:

• Available memory
• Inference backend
• Quantization
• Prompt size
• Conversation history
• Runtime configuration

Safety

Users and application developers are responsible for implementing appropriate safety measures.

Recommended safeguards include:

• Human review
• Output filtering
• Input validation
• Rate limiting
• Privacy protection
• Abuse monitoring
• Clear user disclosures
• Independent fact-checking
• Restricted access for high-risk workflows

Applications using this model should clearly inform users that AI-generated responses may contain errors.

Responsible Use

Do not use Cella Omin 1.1 to facilitate:

• Illegal activity
• Violence or physical harm
• Harassment or abuse
• Fraud or deception
• Impersonation
• Privacy violations
• Unauthorized surveillance
• Malware development
• Exploitation
• Discrimination
• Dangerous or safety-critical decisions

Users are responsible for ensuring that their use of the model complies with applicable laws, regulations, licenses, and platform policies.

Disclaimer

Cella Omin 1.1 is provided as an experimental model.

The model is provided without guarantees of accuracy, reliability, fitness for a particular purpose, or suitability for high-risk applications.

Model outputs do not necessarily represent the views of Qatrix Infotech.

Users are responsible for evaluating the model and its outputs before deployment or use.

Developer

Developed by Qatrix Infotech.

License

This model is released under the Apache License 2.0.

Review the repository license and any applicable dependency licenses before redistribution or commercial use.