yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 28, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2 is a 12 billion parameter Gemma 4 fine-tune by yuxinlu1, specifically optimized for coding and agentic tasks. It excels at multi-step technical tasks, tool use, and debugging, achieving a 3.5x higher score on the tau2-bench telecom agentic benchmark compared to the base model. This model is designed for builders to create custom quants, fine-tune further, or run in transformers, offering strong performance in specialized coding environments.

Loading preview...

Overview

yuxinlu1/gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2 is a 12 billion parameter Gemma 4 fine-tune, focusing on coding and agentic capabilities. This v2 release significantly enhances the model's ability to perform multi-step technical tasks, including reading, reasoning, using tools, and debugging, addressing limitations of its predecessor.

Key Capabilities

  • Agentic Tool Use: Designed for real multi-step tool-use trajectories, mirroring terminal/debugging workflows. It emits structured tool-calls in Gemma 4's native protocol.
  • Enhanced Coding: Incorporates verified chain-of-thought over Python tasks and Fable-5-redo datasets for complex coding scenarios.
  • Grounded Reasoning: Demonstrates 0% fabrication on coding/terminal tasks, consistently grounding its actions by first performing operations like grep, read, or ls.
  • Performance: Achieves approximately 3.5x higher scores on the tau2-bench telecom agentic tool-use benchmark compared to the base gemma-4-12B-it model.

Use Cases

This model is ideal for developers and builders who need a specialized local coding/agentic worker. It is suitable for:

  • Rolling custom quants: Provides un-quantized master weights (bf16) for creating custom GGUF, MLX, AWQ, or GPTQ builds.
  • Further fine-tuning: Serves as a clean base for additional LoRA or continued training.
  • Running in transformers: Compatible with recent transformers builds supporting gemma4_unified architecture.

While highly specialized, it trades some general-knowledge breadth for its coding and agentic strength, making it a focused tool for technical problem-solving.

Finetunes
2 models