adaption_llama_4_defi_wallet_risk_classif_301143e0

Foresight AI DeFi Wallet Risk Classifier

This repository contains the second AutoScientist-trained LoRA adapter for Foresight AI, a finance-category DeFi wallet risk classification project built for the Adaption Labs AutoScientist Challenge.

The model was trained to classify DeFi wallet behavior samples into lower-risk or higher-risk 14-day outcome labels.

Expected output:

• Label 0: lower-risk 14-day outcome
• Label 1: higher-risk 14-day outcome

Training Run

• Run identifier: 301143e0
• Base model: meta-llama/Llama-4-Scout-17B-16E-Instruct
• Export base path in adapter config: togethercomputer/Llama-4-Scout-17B-16E-Instruct_bnb_4bit
• Training method: SFT
• Adapter type: LoRA
• Data format: chat/instruction
• Epochs: 3
• Global steps: 489
• LoRA rank: 32
• LoRA alpha: 64
• Final exported eval loss: 1.072265625

Dataset

This run was trained after expanding the dataset through Adaption Labs with additional domain-specific and general-purpose datapoints.

The task is DeFi wallet risk classification from behavioral feature prompts, including signals such as transaction count, action ratios, asset concentration, protocol concentration, data quality, and wallet profile.

Important: these labels are retrospective proxy labels for hackathon model development. They should not be described as production-verified liquidation, loss, or drawdown events.

Intended Use

This adapter is intended for research and demonstration around DeFi wallet risk classification. It can support decision-support workflows where model output is presented as a risk-review signal, not as financial advice.

Files

• adapter_model.safetensors
• adapter_config.json
• tokenizer.json
• tokenizer_config.json
• special_tokens_map.json
• chat_template.jinja
• trainer_state.json

License And Base Model Terms

Use of this adapter is subject to the license and acceptable-use terms of the base model, Hugging Face, and Adaption Labs challenge requirements.