Overview
Weni/ZeroShot-Llama2-13B-nenhuma is a 13 billion parameter language model built upon the Llama 2 architecture, developed by Weni. This model has been specifically fine-tuned using 20,000 data points in Portuguese, formatted for prompt-based classification tasks. Its core functionality revolves around receiving an input dictionary that contains a phrase to be classified and a set of class options, notably including a 'none' class for instances that don't fit any provided category.
Key Capabilities
- Zero-shot text classification: Classifies Portuguese phrases into categories provided at inference time without prior examples for those specific categories.
- 'None' class handling: Designed to explicitly identify when a phrase does not belong to any of the given class options.
- Portuguese language proficiency: Optimized for understanding and classifying text in Portuguese.
Training Details
The model was trained using bitsandbytes quantization, specifically load_in_4bit: True with bnb_4bit_quant_type: nf4 and bnb_4bit_compute_dtype: bfloat16, indicating an efficient training approach. The training utilized PEFT version 0.4.0.
Good For
- Automated categorization of customer feedback or support tickets in Portuguese.
- Content moderation systems requiring dynamic classification of text.
- Any application needing to sort Portuguese phrases into user-defined categories, including the option to flag irrelevant inputs.