bratao/llama7b-finetuned-openie-lora
The bratao/llama7b-finetuned-openie-lora is a 7 billion parameter Llama-2-based decoder-only causal language model, developed by bratao, specifically fine-tuned for Portuguese generative Open Information Extraction (OpenIE). This full model, not a LoRA adapter, is designed to extract information in ARG0, V, ARG1 format from Portuguese sentences. It serves as a legacy experimental checkpoint for OpenIE tasks in Portuguese.
Loading preview...
Overview
This model, bratao/llama7b-finetuned-openie-lora, is a 7 billion parameter Llama-2-based causal language model. Despite its name, it is a full model (approximately 13.48 GB in PyTorch weight shards), not a small LoRA adapter. It is specifically fine-tuned for Portuguese generative Open Information Extraction (OpenIE), aiming to extract information in the ARG0, V, ARG1 format from given sentences.
Key Characteristics
- Base Architecture: Identifies as
NousResearch/Llama-2-7b-hfwith a Llama decoder-only architecture (32 layers, hidden size 4,096). - Task: Specialized in Portuguese extractive OpenIE.
- Prompting: Uses a specific system instruction for extraction, such as "Dada uma frase S você consegue fazer extrações no formato ARG0 , V, ARG1. Realize a extração para a frase abaixo:" followed by "S: {sentence}".
- Legacy Status: Documented as a legacy experimental checkpoint, with some discrepancies regarding its exact Llama 2/Llama 3 identity and associated evaluation metrics.
Limitations and Considerations
- Model Identity: There are unresolved conflicts regarding its base model identity (Llama 2 vs. Llama 3 as referenced in related thesis work).
- Evaluation: No quantitative metrics can be definitively assigned to this specific public artifact due to checkpoint checksum and identity issues.
- Resource Requirements: Requires approximately 13.48 GB for download and at least 16 GB of free VRAM for unquantized execution.
- License: No explicit license is declared in the public repository; users should clarify licensing with the author and Llama 2 base terms before use or redistribution.