kth8/gemma-3-1b-it-OpenCode-Title-Generator

TEXT GENERATIONConcurrency Cost:1Model Size:1BQuant:BF16Ctx Length:32kPublished:Jun 1, 2026License:gemmaArchitecture:Transformer Cold

The kth8/gemma-3-1b-it-OpenCode-Title-Generator is a 1 billion parameter instruction-tuned language model, fine-tuned from unsloth/gemma-3-1b-it. It specializes in generating concise, relevant titles for conversations, adhering to specific length and content rules. This model is optimized for integration with OpenCode's title agent, providing single-line, grammatically correct titles under 50 characters.

Loading preview...

Model Overview

The kth8/gemma-3-1b-it-OpenCode-Title-Generator is a specialized 1 billion parameter language model, fine-tuned from unsloth/gemma-3-1b-it. Its primary function is to generate brief, descriptive titles for conversations, specifically designed to integrate with OpenCode's title agent.

Key Capabilities

  • Title Generation: Outputs single-line titles, strictly limited to 50 characters, without explanations or additional text.
  • Contextual Relevance: Generates titles that help users quickly find past conversations, focusing on the main topic or question.
  • Rule Adherence: Follows a comprehensive set of rules, including maintaining the original language, ensuring grammatical correctness, avoiding tool names, and handling short or conversational inputs appropriately.
  • Technical Term Preservation: Retains technical terms, numbers, filenames, and HTTP codes in titles.

Training Details

The model was trained using PEFT (Parameter-Efficient Fine-Tuning) with LoRA (Rank: 32, Alpha: 64) on the kth8/title-generation-25000x dataset. It underwent 1 epoch of supervised fine-tuning with a batch size of 8 and a learning rate of 0.0002, achieving a best validation loss of 0.999963. The training utilized an NVIDIA A100-SXM4-40GB GPU, with a peak VRAM usage of 13.854 GB.

Good For

  • Automated Title Creation: Ideal for applications requiring automatic, concise summarization of conversation topics into titles.
  • OpenCode Integration: Specifically configured for use within the OpenCode framework as a small_model for title generation tasks.
  • Structured Output: Suitable for scenarios where strict output format (single line, character limit) is critical.