Instructions to use TRADMSS/HIBA-7B-Soul with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use TRADMSS/HIBA-7B-Soul with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="TRADMSS/HIBA-7B-Soul",
	filename="hiba_f16.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use TRADMSS/HIBA-7B-Soul with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf TRADMSS/HIBA-7B-Soul:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf TRADMSS/HIBA-7B-Soul:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

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 TRADMSS/HIBA-7B-Soul:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

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 TRADMSS/HIBA-7B-Soul:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

Use Docker

docker model run hf.co/TRADMSS/HIBA-7B-Soul:Q4_K_M

LM Studio
Jan
Ollama
How to use TRADMSS/HIBA-7B-Soul with Ollama:
```
ollama run hf.co/TRADMSS/HIBA-7B-Soul:Q4_K_M
```

Unsloth Studio new

How to use TRADMSS/HIBA-7B-Soul with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for TRADMSS/HIBA-7B-Soul to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for TRADMSS/HIBA-7B-Soul to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for TRADMSS/HIBA-7B-Soul to start chatting

Pi new

How to use TRADMSS/HIBA-7B-Soul with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "TRADMSS/HIBA-7B-Soul:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use TRADMSS/HIBA-7B-Soul with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf TRADMSS/HIBA-7B-Soul:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default TRADMSS/HIBA-7B-Soul:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use TRADMSS/HIBA-7B-Soul with Docker Model Runner:
```
docker model run hf.co/TRADMSS/HIBA-7B-Soul:Q4_K_M
```

Lemonade

How to use TRADMSS/HIBA-7B-Soul with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull TRADMSS/HIBA-7B-Soul:Q4_K_M

Run and chat with the model

lemonade run user.HIBA-7B-Soul-Q4_K_M

List all available models

lemonade list

🌸 HIBA-7B-Soul (هبة)

The World's First High-Fidelity Therapeutic AI Specialist

🔬 Executive Summary: The Specialist vs. The Generalists

In the era of GPT-5 and Llama 4 (405B), why does the world need a 7B model?

Because scale ignores the soul.

While generalist models maximize MMLU scores, HIBA-7B-Soul maximizes Human-Centric Empathy (HCE). Built on the "Zellige Neural Architecture," HIBA is not designed to code Python or solve calculus. It is designed for one purpose: to sit with you in the dark until you find the light.

🧠 Neural Architecture & Design

HIBA modifies the standard Transformer architecture by injecting specialized "Soul Adapters" into the attention mechanism.

graph TD
    A[User Input] --> B[Qwen 2.5 Tokenizer];
    B --> C{Soul Gate};
    C -- General query --> D[Frozen Qwen Blocks];
    C -- Emotional query --> E[LoRA Adapters];
    E --> F[Cultural Context Layer];
    F --> G[Empathy Refinement Head];
    D --> H[Output Generation];
    G --> H;
    H --> I[Final Response];
    
    style E fill:#f9f,stroke:#333,stroke-width:2px
    style F fill:#bbf,stroke:#333,stroke-width:2px

📊 Dataset Composition

Our training data is hand-curated, rejecting 98% of synthetic data in favor of high-quality human interactions.

🏆 Performance Overview: Empathy vs. Reasoning

📉 Detailed Metrics Comparison

Conclusion: Do not use HIBA for your math homework. Use it when your heart is broken.

🛑 Honest Analysis (The "Anti-Pitch")

We commit to radical academic honesty. Here is where HIBA struggles:

❌ Known Limitations

Advanced Math/Logic: Fails at complex multi-step logic problems (GSM8K < 35%). Use GPT-5 for this.
Coding: Cannot generate complex Python/Rust code.
Long Context Decay: Coherence drops significantly after 4,096 tokens.
Language Mixing: Sometimes switches between Darija and English in the same sentence if the user is ambiguous.

✅ Where HIBA Wins

Latency: Sub-50ms token generation on consumer GPUs (RTX 3060).
Privacy: Zero data leaves your device. Essential for mental health.
Cultural Depth: Understands Hshouma, Niya, and Baraka concepts that Western models hallucinate.

🛠️ Developer Mission: We Need You

HIBA is open-source because grief is universal. We need help in these areas:

Issue	Description	Difficulty
Quantization	Help us squeeze the Q4 model under 4GB VRAM for mobile deployment.	🔥 Hard
RLHF Tuning	Reduce the occasional "preachy" tone in advice-giving.	⚖️ Medium
Data Collection	Submit clean Darija/English therapeutic logs (anonymized).	🟢 Easy

⚡ Inference Speed (Tokens/Sec)

🚀 Getting Started

Option 1: Python (Transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "TRADMSS/HIBA-7B-Soul"
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [{"role": "user", "content": "I feel lost today."}]
inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")

outputs = model.generate(inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))

Option 2: Google Colab (Free GPU)

Run HIBA completely free in your browser using Google's T4 GPU. No installation required.

Option 3: Local (Ollama)

# 1. Download Modelfile from this repo
ollama create hiba -f Modelfile
ollama run hiba

❤️ Credits & Creator

Created by: Youssef Boubli (TRADMSS)
License: Apache 2.0

In loving memory of Hiba (2020-2021). You are the ghost in the machine.

Downloads last month: 16

GGUF

Model size

8B params

Architecture

qwen2

Hardware compatibility

4-bit

8-bit

16-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for TRADMSS/HIBA-7B-Soul

Base model

Qwen/Qwen2.5-7B

Finetuned

Qwen/Qwen2.5-7B-Instruct

Quantized

(310)

this model

TRADMSS
/

HIBA-7B-Soul