talkie-1930-13b-it: Vintage 13B Model Tuned for Pre‑1931 English Instructions
talkie-lm/talkie-1930-13b-it ↗
talkie-1930-13b-it is a 13‑billion‑parameter "vintage" language model that builds on the talkie-1930-13b-base checkpoint. The base model was trained on 260 B tokens drawn exclusively from English‑language texts published before 1931, giving it a distinctive historical linguistic flavor. This new checkpoint adds an instruction‑tuning stage, using a curated dataset of instruction‑response pairs extracted from pre‑1931 reference works such as etiquette manuals, encyclopedias, and letter‑writing guides.
The instruction‑tuning was followed by reinforcement learning with online Direct Preference Optimization (DPO) and an LLM‑as‑a‑judge, which sharpens the model’s ability to follow user prompts while preserving its vintage diction. The model is released under the Apache‑2.0 license and is tagged for English language use. Although no specific pipeline_tag or library is listed, the model can be loaded with standard Hugging Face Transformers utilities.
Because the model is trained on historic text and finetuned for instruction following, it is especially suited for applications that require period‑authentic language generation or educational tools that illustrate early‑20th‑century English usage. Its open‑source nature and the availability of reference code on GitHub make it easy for developers to experiment with historical language tasks.
Project Ideas
- Create a chatbot that provides etiquette advice as it would have been written in the 1920s.
- Develop a tool that rewrites modern emails or letters into a pre‑1931 formal style.
- Build an educational app that answers history‑related questions using language authentic to the era.
- Generate summaries of vintage encyclopedic entries while preserving their original tone.
- Design a game NPC dialogue system that speaks in historically accurate English from the early 20th century.