---
language:
- en
license: other
license_name: s2-mas-controlled-research-v1.2
license_link: https://huggingface.co/ya9919/fin-laya-3-rounds/blob/main/LICENSE
base_model: convaiinnovations/laya
base_model_relation: finetune
tags:
- s2-mas
- financial-one-shot
- decision-model
---

# FinL · three-round comparator (native PyTorch)

## Model naming and experiment labels

This is the **FinL three-round comparator**, in native Laya/PyTorch format. The main **FinL** name refers to the [full six-round checkpoint](https://huggingface.co/ya9919/FinL). Its paired earlier MLX format remains [fin-laya-mlx-3-rounds](https://huggingface.co/ya9919/fin-laya-mlx-3-rounds).

## 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**. These findings connect individual choices to consumption, cash and payment paths over time.

## Repository role and training

This is the **native FinL 3-round financial checkpoint**, with approximately 421M parameters, based on Laya's ModernBERT-large encoder and option-scoring head. Two SFT stages followed by native RLCD adapt the model to real-survey data, using **1,721,623 training observations per language**. The [MLX runtime](https://huggingface.co/ya9919/fin-laya-mlx-3-rounds) packages this trained checkpoint for FP16 deployment. The [6-round continuation](https://huggingface.co/ya9919/FinL) extends its supervised task coverage.

## 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 | 74.5336% | 38.0164% |
| Expanded test | English | PyTorch BF16 | 418,597 | 862 | 56.3657% | 19.2395% |

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

On the same shared test, original Laya in native BF16 scores 60.2080% accuracy and 29.5574% F1; financial training raises these to the 3-round results above. See Figure 5 and Methods M4/M6.

## Deployment timing

The manuscript's single-request timing uses the [FinL-bilingual checkpoint](https://huggingface.co/ya9919/fin-laya-multilingual-mlx) 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.

## License and manuscript

The project model contribution is offered under the [Controlled Model Research License v1.2](LICENSE). 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](https://github.com/KataDavidXD/s2-mas/blob/main/paper/S2_MAS_Paper.pdf) · [Project](https://github.com/KataDavidXD/s2-mas)

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.