super-inference/super-12B-it-ste

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 14, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

The super-inference/super-12B-it-ste is a 12 billion parameter instruction-tuned model, likely based on the Gemma-4 architecture, optimized for efficient deployment and inference. This model is provided in GGUF format, enabling faster processing and reduced memory footprint, particularly beneficial for local or edge device applications. Its primary strength lies in its optimized GGUF conversion, making it suitable for various text-based generative AI tasks where resource efficiency is critical. The model leverages Unsloth for accelerated finetuning and conversion.

Loading preview...

Model Overview

The super-inference/super-12B-it-ste is a 12 billion parameter instruction-tuned language model, specifically converted to the GGUF format for optimized inference. This model appears to be based on the gemma-4-12B-agentic-fable5-composer2.5-v2 architecture, indicating a focus on agentic capabilities and composition.

Key Characteristics

  • GGUF Format: Provided in GGUF format, which is highly optimized for CPU inference and compatibility with llama.cpp and related tools, enabling efficient deployment on a wide range of hardware.
  • Unsloth Optimization: The model was finetuned and converted using Unsloth, a library known for accelerating the finetuning process by up to 2x.
  • Available Quantizations: Includes Q8_0.gguf for balanced performance and size, and BF16-mmproj.gguf which suggests potential for multimodal capabilities, although the primary README focuses on text-only usage.

Use Cases

This model is particularly well-suited for:

  • Efficient Local Inference: Ideal for running generative AI applications on consumer-grade hardware or edge devices due to its GGUF optimization.
  • Instruction-Following Tasks: As an instruction-tuned model, it is designed to follow user prompts and generate relevant text outputs.
  • Rapid Prototyping: The Unsloth-optimized finetuning process implies it can be quickly adapted or used for various text generation tasks.