TensorVizion/Loi-LLM

Loi-LLM: A Fine-Tuned Conversational Assistant

Loi-LLM, developed by TensorVizion, is a 3.2 billion parameter conversational language model built upon Meta's Llama 3.2 3B Instruct. It has been fine-tuned using QLoRA (4-bit quantized low-rank adaptation) to enhance its performance in general chat and assistant tasks.

Key Capabilities & Features

• Conversational Coherence: Improved ability to maintain natural and logical dialogue flows.
• Instruction Following: Enhanced accuracy in understanding and executing user instructions.
• Response Quality: Delivers higher quality and more relevant responses for general chat.
• Efficient Deployment: Available in various formats, including full fp16, Q6_K GGUF, and Q4_K_M GGUF quantizations for flexible local inference with tools like llama.cpp or LM Studio.
• PEFT Adapter: Can be loaded as a PEFT LoRA adapter on top of the base Llama 3.2 3B Instruct model for easy integration.

Intended Use Cases & Limitations

Loi-LLM is primarily designed for:

• General chat applications.
• Assistant tasks requiring conversational AI.

Limitations:

• English Only: Performance in other languages is not guaranteed.
• Scale-Dependent Reasoning: As a 3B parameter model, it may be outperformed by larger models on complex reasoning tasks.
• Specialized Tasks: Not specifically trained for code generation, advanced mathematics, or domain-specific professional tasks.
• Hallucination Risk: Like all LLMs, it can produce inaccurate or fabricated information.
• Safety: Not aligned for safety-critical or high-stakes applications.