Overview
cenfis/llama3-8b-tr-finetuned is an experimental and educational fine-tuned variant of the 8 billion parameter Llama 3 model, specifically adapted for the Turkish language. This model was developed by cenfis, building upon Unsloth's Llama 3-8B quantized model using an adapter-based approach. The project aims to provide a Llama 3 model with enhanced proficiency in Turkish, with plans for continuous development and training as more quality Turkish data becomes available.
Key Capabilities
- Turkish Language Processing: Fine-tuned to understand and generate text in Turkish.
- Llama 3 Architecture: Benefits from the underlying Llama 3 8B parameter architecture.
- Adapter-based Fine-tuning: Utilizes PEFT (Parameter-Efficient Fine-Tuning) for efficient adaptation.
- Quantized Base Model: Built on a 4-bit quantized Llama 3-8B model, potentially offering efficiency benefits.
- Format Availability: Provided in .gguf and .bin formats for compatibility with
llama.cpp and vLLM.
Important Usage Notes
- Prompt Template: It is highly recommended to use an Alpaca Prompt Template or a similar structured template to ensure meaningful generations and avoid repetitive outputs.
- Hardware Requirement: Requires a CUDA-supported GPU for optimal performance.
- Development Status: This is an experimental and educational model, open to further development and improvement with additional quality Turkish datasets.