Instructions to use muhamedkamil/berna-r7-mother with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use muhamedkamil/berna-r7-mother with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="muhamedkamil/berna-r7-mother")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("muhamedkamil/berna-r7-mother", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use muhamedkamil/berna-r7-mother with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "muhamedkamil/berna-r7-mother"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "muhamedkamil/berna-r7-mother",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/muhamedkamil/berna-r7-mother

SGLang

How to use muhamedkamil/berna-r7-mother with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "muhamedkamil/berna-r7-mother" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "muhamedkamil/berna-r7-mother",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "muhamedkamil/berna-r7-mother" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "muhamedkamil/berna-r7-mother",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use muhamedkamil/berna-r7-mother with Docker Model Runner:
```
docker model run hf.co/muhamedkamil/berna-r7-mother
```

Berna R7 - Root Model

A 1.06B multilingual language model with a modular cell-based architecture.

This repository will host the Root model of Berna R7, the first large-scale implementation of the DNA-Kernel Plexus architecture from Berna R5.

Status: Training in progress. Weights will be uploaded on or around 5 October 2026.

Zenodo DOI: 10.5281/zenodo.23037065 Mendeley DOI: 10.17632/k24wwx6tp6.1

Model Details

Property	Value
Parameters	1,056,265,728 (1.06B)
Architecture	Decoder-only Transformer (Berna)
Layers	28
Hidden size	1536
Intermediate size	6144
Attention heads	16 (GQA, 4 KV heads)
Vocab size	32,000 (BPE)
Context length	4,096
RoPE theta	500,000
Normalization	RMSNorm
Activation	SwiGLU
Precision	BF16
Tied embeddings	Yes

Training

Property	Value
Data	2.66B tokens (English + Math + Code)
Epochs	1
Hardware	1x RTX 5090 (32 GB)
Duration	~6.5 days
Optimizer	AdamW8bit
Learning rate	3e-4 (cosine, 500-step warmup)
Batch size	64 x 4096 = 262,144 tokens/step
Gradient checkpointing	enabled

Architecture (Berna R5 implementation)

Six-layer structure from R5:

Sensors -> Spinal Cord -> Plexus -> Knowledge Cells -> DNA Kernel -> Registry

Knowledge Cells: 112 cells (4 per MLP layer x 28 layers).
6D Knowledge Vector: K = [L, W, H, D, T, E], saturation S = ||K||_omega.
DNA Kernel: 100 chromosomes x 4 genes (a, s, p, pi).
Plexus: dynamic graph with edges updated from co-activations.
Registry: SQL catalog with UUIDs and connection tokens.

Compliance with R5

R5 Component	Status
Registry + UUIDs	implemented
connection_token	implemented
Federation	implemented
Knowledge Cells	112 cells
6D K vector	K1, K3, K5, K6
DNA Kernel	100x4
S* = 0.85	yes
Splitting (H1)	tested
Plexus	implemented
Router (Spinal Cord)	deferred
Bounded Forgetting (H2)	not yet tested

Intended Use

Research on continual learning and growing neural networks.
Base model for fine-tuning on domain-specific tasks.
Reference implementation of the R5 DNA-Kernel Plexus framework.

Out-of-Scope Use

Production deployment without evaluation.
Safety-critical applications.
Uses prohibited by the Berna Research License v1.0.

Limitations

Undertrained: 1 epoch over 2.66B tokens (~2.5 tokens/param).
Text-only: no vision, no audio.
Not instruction-tuned.
No benchmarks yet (MMLU, GSM8K, HumanEval pending).
English + Math + Code only.

What This Model Does NOT Claim

State-of-the-art on any benchmark.
Superiority over GPT-4-class models.
Scalability beyond 1.5B parameters.
Theoretical completeness.

Citation

@software{berna_r7_2026,
  author    = {Muhammed, Mohammed Kamil},
  title     = {Berna R7: A 1.06B Multilingual Language Model},
  year      = {2026},
  publisher = {Berna Labs},
  url       = {https://github.com/Berna-Labs/berna-r7}
}

For the underlying theory:

@software{berna_r5_2026,
  author    = {Muhammed, Mohammed Kamil},
  title     = {DNA-Kernel Plexus},
  year      = {2026},
  doi       = {10.5281/zenodo.23015308}
}

License

Berna Research License v1.0. Research use is free. Commercial use requires a separate license. Contact: info@bernalabs.com

Downloads last month: -; Downloads are not tracked for this model. How to track