Qwen3_negotiation-assistant-united_nations

This model is a fine-tuned version of unsloth/qwen3-14b-unsloth-bnb-4bit using Unsloth and the Hugging Face TRL library for accelerated training.

Model Description

• Base model: unsloth/qwen3-14b-unsloth-bnb-4bit
• Architecture: Qwen3-14B (optimized with Unsloth)
• Fine-tuned by: ilyass31
• License: Apache 2.0
• Language: English
• Precision: 4-bit quantization using bnb for efficient inference

Use Case

This model is designed as an AI negotiation assistant, particularly for domains such as:

• Humanitarian negotiations
• Climate diplomacy
• Stakeholder mapping
• Mediation scenarios involving multi-party interests

It can generate:

• Islands of Agreement
• Stakeholder Influence Maps
• Negotiation strategies and recommendations

Model Training

This model was fine-tuned using:

-QLoRA / LoRA adapters for efficient fine-tuning. -4-bit quantized base model for memory efficiency during inference. -Supervised fine-tuning using negotiation-based prompts and domain-specific responses.

Training Hyperparameters

-Optimizer: AdamW -Learning rate: 5e-5 -Batch size: 16 -Epochs: 3 -Warmup ratio: 0.1

Limitations

-GPU Requirement: This model relies on GPU hardware, as Unsloth only supports CUDA devices.

Requirement:

Install bitsandbytes using !pip install bitsandbytes before calling the model.

Citation

If you use this model in your research or application, please cite the following: @misc{ilyass31_dh_ai_negotiation_assistant_2025,

author = {ilyas DAHAOUI},

title = {Qwen3_negotiation-assistant-united_nations},

year = {2025},

url = {https://huggingface.co/ilyass31/DH-AI-negotiation-assistant},

note = {Accessed: 2025-05-03} }