chanifrusydi/gemma-3-270m-lora-finetuned

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.3BQuant:BF16Context Size:32kPublished:Jan 22, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

chanifrusydi/gemma-3-270m-lora-finetuned is a 0.3 billion parameter Gemma 3 model, fine-tuned with LoRA adapters on approximately 100,000 instruction-following samples from the mlabonne/FineTome-100k dataset. This model is optimized for instruction following, conversational AI, question answering, and text generation tasks. It supports a context length of 32768 tokens and is available in Safetensors and GGUF formats for various deployment scenarios.

Loading preview...

Overview

This model, chanifrusydi/gemma-3-270m-lora-finetuned, is a compact yet capable language model based on the Gemma 3 270M architecture. It has been fine-tuned using LoRA (Low-Rank Adaptation) adapters on a substantial instruction-following dataset, mlabonne/FineTome-100k, comprising around 100,000 samples. The fine-tuning process utilized the Unsloth framework for efficiency, completing in approximately 98 minutes on an NVIDIA RTX 5070 Ti.

Key Capabilities

  • Instruction Following: Designed to accurately follow user instructions.
  • Conversational AI: Suitable for developing chatbots and interactive agents.
  • Question Answering: Can provide answers to queries based on its training.
  • Text Generation: Capable of generating coherent and contextually relevant text.

Technical Details

  • Base Model: unsloth/gemma-3-270m-it
  • Parameters: 0.3 billion
  • Context Length: 32768 tokens
  • Training Data: mlabonne/FineTome-100k (100k samples)
  • LoRA Configuration: r=128, alpha=128, targeting q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj modules.
  • Available Formats: Safetensors (for Transformers, vLLM) and GGUF (Q8_0, BF16 for llama.cpp, Ollama).

Usage Considerations

This model is well-suited for applications requiring a small, efficient instruction-tuned model. However, it is explicitly noted as out-of-scope for generating medical or legal advice, creating harmful content, or production use without thorough evaluation.