voidful/gemma-4-e4b-it-agent-sft-tw

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 22, 2026License:gemmaArchitecture:Transformer Featherless Exclusive Cold

The voidful/gemma-4-e4b-it-agent-sft-tw is a 7.9 billion parameter Gemma-4-E4B-it model, developed by voidful, that has undergone a true full fine-tune on the voidful/agent-sft dataset. This model is specifically optimized for agent-based tasks and evaluated against the voidful/claw-eval-zh benchmark for traditional Chinese language performance. It is designed for applications requiring robust agentic capabilities in a traditional Chinese context.

Loading preview...

Model Overview

This model, voidful/gemma-4-e4b-it-agent-sft-tw, is a 7.9 billion parameter language model based on the google/gemma-4-E4B-it architecture. It has been subjected to a true full fine-tune on the voidful/agent-sft dataset, distinguishing it from LoRA, QLoRA, or merged adapter approaches. The training configuration confirms a full fine-tune without 4-bit or 8-bit quantization of the base model.

Key Characteristics

  • Full Fine-Tune: Unlike many models that use parameter-efficient fine-tuning (PEFT) methods, this model is a complete fine-tune of the base Gemma-4-E4B-it model.
  • Agentic Optimization: The fine-tuning on voidful/agent-sft suggests a specialization in agent-based tasks and interactions.
  • Traditional Chinese Evaluation: The model's performance was specifically evaluated using voidful/claw-eval-zh with the --language tw flag, indicating an optimization and assessment for traditional Chinese language contexts.
  • Performance: Achieved a mean score of 0.4789416666666667 (9.578833333333334 / 20) on the traditional Chinese evaluation, with google/gemma-4-31B-it serving as the judge model.

Use Cases

This model is particularly well-suited for applications requiring:

  • Agent-based systems: Where the model needs to perform actions, reason, or interact in a structured manner.
  • Traditional Chinese language processing: For tasks and applications targeting users or content in traditional Chinese.
  • Research and development: For exploring the impact of full fine-tuning on Gemma-4-E4B-it for agentic capabilities and specific language adaptations.

Finetunes
1 models