agaralon/talkie-1930-13b-base-vllm
The agaralon/talkie-1930-13b-base-vllm is a 13 billion parameter base language model, repackaged for vLLM and HuggingFace compatibility. Developed by talkie-lm, it was pretrained on approximately 260 billion tokens of English text published before 1931. This model is designed for text completion tasks, generating continuations based on provided prompts. Its unique training data makes it specialized for generating text in the style and vocabulary of the pre-1931 era.
Loading preview...
Model Overview
agaralon/talkie-1930-13b-base-vllm is a 13 billion parameter base language model, a repackaging of the original talkie-lm/talkie-1930-13b-base for seamless integration with HuggingFace transformers and vLLM serving. The original model lacked standard HuggingFace configurations, which this repackage addresses by providing config.json, tokenizer.json, and modeling_talkie.py.
Key Characteristics
- Historical Text Generation: Pretrained on approximately 260 billion tokens of English text published before 1931, making it uniquely suited for generating content in a historical style.
- Base Model: This is a completion model, not instruction-tuned, and does not use a chat template. It excels at continuing raw text prompts.
- vLLM Compatibility: Includes necessary configurations and modeling code for efficient serving with vLLM, supporting
bfloat16inference. - Context Length: Features a context window of 4096 tokens.
- Sampling Recommendation: For optimal results, use a
temperatureof 0.5 or higher, as greedy decoding (temperature=0) can lead to repetitive outputs.
Use Cases
- Historical Content Creation: Generating narratives, dialogues, or documents that mimic the linguistic style of the pre-1931 era.
- Research: Studying language patterns and evolution by generating text based on historical corpora.
- Creative Writing: Assisting authors in crafting period-accurate prose for historical fiction or similar projects.