Instructions to use Skebobic/Bobic-1.6-Raye with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Skebobic/Bobic-1.6-Raye with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Skebobic/Bobic-1.6-Raye:Q8_0
# Run inference directly in the terminal:
llama cli -hf Skebobic/Bobic-1.6-Raye:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Skebobic/Bobic-1.6-Raye:Q8_0
# Run inference directly in the terminal:
llama cli -hf Skebobic/Bobic-1.6-Raye:Q8_0

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Skebobic/Bobic-1.6-Raye:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf Skebobic/Bobic-1.6-Raye:Q8_0

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Skebobic/Bobic-1.6-Raye:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Skebobic/Bobic-1.6-Raye:Q8_0

Use Docker

docker model run hf.co/Skebobic/Bobic-1.6-Raye:Q8_0

LM Studio
Jan

vLLM

How to use Skebobic/Bobic-1.6-Raye with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Skebobic/Bobic-1.6-Raye"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Skebobic/Bobic-1.6-Raye",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Skebobic/Bobic-1.6-Raye:Q8_0

Ollama
How to use Skebobic/Bobic-1.6-Raye with Ollama:
```
ollama run hf.co/Skebobic/Bobic-1.6-Raye:Q8_0
```
Unsloth Desktop
Docker Model Runner
How to use Skebobic/Bobic-1.6-Raye with Docker Model Runner:
```
docker model run hf.co/Skebobic/Bobic-1.6-Raye:Q8_0
```

Lemonade

How to use Skebobic/Bobic-1.6-Raye with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Skebobic/Bobic-1.6-Raye:Q8_0

Run and chat with the model

lemonade run user.Bobic-1.6-Raye-Q8_0

List all available models

lemonade list

Atomic Chat

Bobic 1.6 Raye (125.86M Parameters)

Bobic 1.6 Raye is an optimized, RL/GRPO policy-aligned Small Language Model (SLM) trained from scratch. It features Grouped-Query Attention (GQA 12:4), SwiGLU feed-forward networks, and a dedicated token-level NumberHead for numerical stability.

Benchmark & Capabilities

MMLU-Pro Zero-shot Accuracy: Reaches 18.33% - 20.00% under calibrated inference.
Hallucination Reductions: Grounded arithmetic (+2=4$, +5=10$, 0-4=6$), commonsense logic, and negation awareness.
Quantization: High-fidelity Q8_0 GGUF included for sub-130MB deployment in LM Studio and llama.cpp.

Model Details

Architecture: Bobic Raye (GQA 12:4, SwiGLU, RMSNorm, Rotary Embeddings)
Parameters: 125,861,376 (~125.86M)
Context Length: 2048 tokens
Vocabulary: 16,384 BPE tokens ()
Included Files:
- : PyTorch FP32 state dictionary & config
- : 8-bit quantized GGUF file (~127.8 MB)
- : Hugging Face BPE Tokenizer definition

Recommended Inference Parameters

For dialogue and reasoning:

Temperature: 0.25 - 0.35
Top-p: 0.85
Repetition Penalty: 1.15
Max Tokens: 128 - 256

Downloads last month: 9

GGUF

Model size

0.1B params

Architecture

bobic

Hardware compatibility

8-bit