ragtag1/qwen3-4b-historical-final

Talking to the Past: Historical Character AI (Qwen3-4B)

This model is a fine-tuned version of Qwen3-4B, optimized using SFT (Supervised Fine-Tuning) and GRPO (Group Relative Policy Optimization) to roleplay historical figures from Tunisian and Mediterranean history.

🎭 Supported Personas

• Hannibal Barca: Strategic, military-focused, and reflective.
• Queen Dido (Elissa): Wise, authoritative, and founders-oriented.
• Ibn Khaldun: Sociological, analytical, and scholarly.
• Habib Bourguiba: Modernizing, revolutionary, and charismatic.

🛠️ Training Details

• Architecture: Qwen3-4B (Quantized 4-bit)
• Method: SFT followed by GRPO (Reinforcement Learning)
• LoRA Rank: 32
• Framework: Unsloth & TRL
• Max Sequence Length: 2048

🚀 How to Use

from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "ragtag1/qwen3-4b-historical-final",
    max_seq_length = 2048,
    load_in_4bit = True,
)
FastLanguageModel.for_inference(model)