ArunSaini95/phi_finetune_4bit
TEXT GENERATIONConcurrency Cost:1Model Size:3.8BQuant:BF16Ctx Length:32kPublished:May 31, 2026License:apache-2.0Architecture:Transformer Open Weights Cold
ArunSaini95/phi_finetune_4bit is a 3.8 billion parameter Phi-3-mini-instruct causal language model, fine-tuned by ArunSaini95. This model was optimized for faster training using Unsloth and Huggingface's TRL library. It is designed for general instruction-following tasks, leveraging its efficient training methodology.
Loading preview...
Overview
ArunSaini95/phi_finetune_4bit is a 3.8 billion parameter language model, fine-tuned from the unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit base model. Developed by ArunSaini95, this model leverages the Unsloth library in conjunction with Huggingface's TRL library to achieve significantly faster training times.
Key Capabilities
- Efficient Training: The model was trained approximately 2x faster due to the integration of Unsloth, which optimizes the fine-tuning process.
- Instruction Following: As a fine-tuned instruct model, it is designed to understand and respond to user instructions effectively.
- Quantized Performance: Built upon a 4-bit quantized base model, it offers a balance between performance and reduced memory footprint.
Good For
- Resource-Constrained Environments: Its 4-bit quantization makes it suitable for deployment where memory and computational resources are limited.
- Rapid Prototyping: The faster training methodology allows for quicker iteration and experimentation with fine-tuning tasks.
- General Purpose Instruction-Following: Can be applied to a variety of tasks requiring the model to follow specific commands or prompts.