MightyOctopus/pricer-merged-model-A-v1
MightyOctopus/pricer-merged-model-A-v1 is an 8 billion parameter LLaMA 3.1-based causal language model, developed by MyungHwan Hong, specifically specialized for numeric price prediction from product text. This model was created by merging LLaMA 3.1 8B with a LoRA adapter (Pricer LoRA v1). It excels at estimating approximate consumer product prices from textual metadata and serves as a base for further LoRA fine-tuning or research into LLM-based numeric regression.
Loading preview...
Model Overview
MightyOctopus/pricer-merged-model-A-v1 is an 8 billion parameter causal language model, developed by MyungHwan Hong, built upon the LLaMA 3.1 architecture. This model is a merged checkpoint, combining the LLaMA 3.1 8B base with a specialized LoRA adapter (Pricer LoRA v1), and is optimized for numeric price prediction.
Key Capabilities
- Numeric Price Prediction: Specialized in estimating approximate consumer product prices based on textual metadata such as product title, description, and category.
- Base for Fine-tuning: Designed to serve as a foundational model for further LoRA fine-tuning, allowing for adaptation to specific domains or datasets.
- Research Tool: Useful for research into LLM-based numeric regression and comparative studies against classical machine learning regressors.
Intended Use Cases
- Base Checkpoint: Direct inference for price prediction or as a starting point for custom LoRA fine-tuning.
- Domain-Specific Pricing: Development of more specialized pricing models.
- Educational Experiments: Exploring LoRA merging strategies and LLM capabilities in numeric tasks.
Limitations and Considerations
- Approximate Predictions: Outputs are estimates and not exact or authoritative price sources.
- Bias: May reflect historical and dataset-specific price biases and is not robust to rapid market changes.
- Out-of-Scope: Not suitable for real-time production pricing systems, financial decision-making, or safety-critical applications. Users should validate predictions with real pricing data and avoid high-stakes commercial deployment.