3tic/Orion-Qwen3-1.7B-CPT-v2604

TEXT GENERATIONConcurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Apr 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

3tic/Orion-Qwen3-1.7B-CPT-v2604 is a 2 billion parameter base model built upon the Qwen3-1.7B architecture, developed by 3tic. It has undergone further training (CPT) on over 40 billion tokens of Japanese and Chinese light novel data. This model is specifically designed for subsequent fine-tuning tasks, excelling in applications requiring understanding and generation of content related to light novels and similar narrative styles in both languages.

Loading preview...

Model Overview

3tic/Orion-Qwen3-1.7B-CPT-v2604 is a 2 billion parameter base model, derived from the Qwen3-1.7B architecture. Its core differentiator lies in its extensive Continued Pre-Training (CPT) on a massive dataset exceeding 40 billion tokens, specifically curated from Japanese and Chinese light novel content. This specialized training makes it particularly adept at processing and generating text within this domain.

Key Training Data

The model's CPT dataset is rich and diverse, encompassing various forms of narrative content in both Japanese and Chinese:

  • Japanese Sources: Published light novel texts, web-based fiction, Galgame scripts, and anime subtitle texts.
  • Chinese Sources: Translated light novel texts, online fiction platforms, light novel forum translations, localized Galgame scripts, translated anime subtitle texts, and certain fanfiction/unofficial site data.

Intended Use

This model is primarily intended as a robust base for further fine-tuning. Its deep understanding of light novel conventions, character interactions, and narrative structures in both languages makes it an excellent foundation for tasks such as:

  • Narrative Generation: Creating new stories or continuations in the light novel style.
  • Character Dialogue: Generating contextually appropriate and genre-specific character dialogue.
  • Translation & Localization: Assisting with translation or localization of light novel content between Japanese and Chinese.
  • Content Analysis: Understanding themes, tropes, and stylistic elements within light novel texts.

Finetunes
1 models