FinL · compatibility entry (six-round; native PyTorch)

Model naming and experiment labels

The canonical main-model repository is ya9919/FinL, the explicit full six-round native checkpoint. ya9919/fin-laya retains its existing weight-bearing compatibility entry and points to FinL. FinL-mlx is the primary MLX runtime; FinQ is the separate Qwen3.5-4B-Base LoRA-plus-head main route.

Project background

S²-MAS is a society-scale multi-agent simulation framework for studying how heterogeneous household decisions become population outcomes under changing conditions. It connects a constructed resource of 500 million household personas, reusable scenarios, trainable decision models and stateful world updates. FinL supplies a reusable decision policy: it scores the complete alternatives for a household state and question, allowing the same policy interface to serve different tasks.

The manuscript's paired-condition application evaluates 12 million distinct synthetic household agents and 24 million choices. A 24-month extension follows 10,000 households through changing income, expenses and bank offers. The applications show how household resources shape consumption responses: persistent income loss draws down reserves and delays purchases, while sustained income gains support both cash reserves and purchases. Repayment feedback affects later offers. These findings connect individual choices to consumption, cash and payment paths over time.

Repository role and training

This is the compatibility entry for the English-base FinL 6-round checkpoint, provided in native Laya/PyTorch format. The named native release and MLX runtime refer to the same trained checkpoint family; this entry does not add a training variant.

The approximately 421M-parameter model adapts Laya's ModernBERT-large encoder and option-scoring head to real-survey data. Its 6-round continuation adds SFT and RLCD over 2,925,263 training examples per language. This is the largest disclosed domain post-training set among the financial decision models compared in the manuscript.

Paper results for this checkpoint

Test	Input language	Backend / precision	QA denominator	Question versions	Accuracy	Question-macro F1
Shared financial questions	English	PyTorch BF16	240,458	119	74.5586%	38.40%
Expanded test	English	PyTorch BF16	418,597	862	67.2728%	27.8355%

Accuracy weights QA records equally. Question-macro F1 averages F1 over the complete candidate classes within each question version, then weights question versions equally.

Native shared-test F1 is shown at the precision displayed in Figure 5; Methods M6 reports the expanded test. The same checkpoint family in MLX FP16 scores 74.5544% accuracy and 38.3746% F1 on the shared test; see the MLX card and Methods M5.

Deployment timing

The manuscript's single-request timing uses the FinL-bilingual checkpoint on one NVIDIA A100 PCIe 40GB in FP16, with batch size and concurrency one and three fixed Chinese cases. Its card and Methods M9 report latency, throughput and the Jev comparison.

Model files and inference

This repository contains model weights, the tokenizer and the inference configuration. Native Laya / PyTorch inference uses BF16. predict_example.py uses an independently generated synthetic example.

This is a compatibility entry; the named native six-round release and MLX six-round release provide their corresponding runtime formats.

License and manuscript

The project model contribution is offered under the Controlled Model Research License v1.2. The full license and retained upstream Apache-2.0 rights are in LICENSE. Training and evaluation records are not distributed with the model.

S²-MAS manuscript · Project

The public project provides scientific plotting, decision-model training and evaluation, and independent synthetic companion-data generation scripts. The GENESE/MAS engine implementation remains private.