uabali/gemma4-e4b-TR

gemma-4-e4b-rag-TR : GGUF

Turkish RAG-optimized GGUF version of the LoRA fine-tuned Gemma 4 E4B model.

This model was fine-tuned with LoRA (QLoRA) on the Turkish RAG dataset and converted to GGUF format using Unsloth.

Model Description

• Base Model: google/gemma-4-e4b
• Fine-tuning: LoRA on Turkish RAG data
• Dataset: Metin/WikiRAG-TR
• Purpose: High-quality Turkish Retrieval-Augmented Generation (RAG)
• Context Length: 8192 tokens

Example Usage (llama.cpp)

# Text-only inference
./llama-cli -hf uabali/gemma4-e4b-rag-TR -p "Soru: ..." --jinja -c 8192

# With server (recommended for multiple users)
./llama-server -hf uabali/gemma4-e4b-rag-TR --port 8080 --cont-batching -np 6 -c 8192