VertexAIco/amethyst-1-mini

VISIONConcurrent Unit Cost:1Model Size:4.3BQuant:BF16Context Size:32kPublished:Aug 7, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

VertexAIco/amethyst-1-mini is a 4.3 billion parameter general-purpose chat and instruction-following model, fine-tuned from Google's Gemma 3 4B IT. Developed as an independent research project, it validates an end-to-end distillation and fine-tuning pipeline on consumer hardware. This model excels as a lightweight conversational assistant for experimentation and research into small-scale distillation methods.

Loading preview...

Amethyst 1 Mini: A Distilled Gemma 3 4B IT Model

Amethyst 1 Mini is an independent research project focused on validating a distillation and fine-tuning pipeline on consumer hardware. This 4.3 billion parameter model is a LoRA fine-tune of Google's Gemma 3 4B IT, specifically using the mlx-community/gemma-3-4b-it-qat-4bit checkpoint.

Key Capabilities

  • General-purpose conversational ability: Fine-tuned on 1,122 synthetically generated instruction/response pairs, distilled from nvidia/nemotron-3-super-120b-a12b.
  • Broad task coverage: The training dataset spans diverse categories including explanation, reasoning, code generation, extraction, planning, roleplay, creative writing, translation, sentiment classification, and brainstorming.
  • Lightweight deployment: The LoRA adapter is fused into the base weights and dequantized to fp16, allowing direct loading with transformers without MLX or quantization dependencies.

Intended Use

Amethyst 1 Mini is designed as a lightweight, general-purpose conversational assistant for:

  • Experimentation with small language models.
  • Research into distillation pipelines.
  • Hobbyist deployment scenarios.

It is important to note that due to its small, synthetic training dataset, the model's behavior can be inconsistent outside its training categories, and it is not intended for high-stakes or production use.