jordanmatsumoto/pricing-specialist

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>

pricing-specialist

This model is a fine-tuned version of meta-llama/Meta-Llama-3.1-8B on pricing_ai dataset. Third and final run — previous two training attempts were interrupted before completion.

Model description

• —Model Type: Fine-tuned LLM for structured price estimation
• —Architecture: Adapted from OpenAI GPT and trained on curated product datasets
• —Primary Use: Predicting item prices from text-based descriptions
• —Integration: Deployed remotely through Modal for scalable inference
• —Experiment Tracking: Managed via Weights & Biases
• —Owner Project: Pricely

Intended uses & limitations

Intended uses

• —Estimating product prices from e-commerce deal descriptions
• —Supporting multi-agent workflows for deal discovery and ranking
• —Demonstrating fine-tuned regression-like behavior in large language models

Limitations

• —Predictions are approximate and may not reflect real-world prices
• —Model performance decreases with ambiguous or incomplete inputs
• —Primarily intended for research and demonstration, not production valuation

Training and evaluation data

• —Dataset: `jordanmatsumoto/pricing_data`
• —Size: ~350,000 structured product entries (title, description, true price)
• —Coverage: Balanced across electronics, appliances, home, and consumer goods
• —Preparation: Cleaned and standardized for fine-tuning

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 2e-05
• —trainbatchsize: 16
• —evalbatchsize: 1
• —seed: 42
• —optimizer: Use OptimizerNames.PAGEDADAMW with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: cosine
• —lrschedulerwarmup_ratio: 0.03
• —num_epochs: 0.25

Framework versions

• —PEFT 0.14.0
• —Transformers 4.48.3
• —Pytorch 2.5.1+cu124
• —Datasets 3.2.0
• —Tokenizers 0.21.4