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

# FinL-mlx · three-round comparator (MLX FP16)

## Model naming and experiment labels

This repository packages the **FinL three-round comparator in MLX FP16**. The main **FinL-mlx** name refers to the [full six-round MLX runtime](https://huggingface.co/ya9919/FinL-mlx). The earlier native counterpart remains [fin-laya-3-rounds](https://huggingface.co/ya9919/fin-laya-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 **MLX FP16 runtime of the FinL 3-round financial checkpoint**. It is paired with the [native release](https://huggingface.co/ya9919/fin-laya-3-rounds), which adapts Laya's approximately 421M-parameter ModernBERT-large encoder and option-scoring head to real-survey data through two SFT stages and native RLCD. The training partition contains **1,721,623 observations per language**. The MLX package provides another runtime for that trained checkpoint.

## Paper results for this checkpoint

| Test | Input language | Backend / precision | QA denominator | Question versions | Accuracy | Question-macro F1 |
|---|---|---|---:|---:|---:|---:|
| Shared financial questions | English | MLX FP16 | 240,458 | 119 | 74.53% | 38.01% |

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

Scores use the two-decimal precision displayed in Supplementary Figure S2. The [paired native card](https://huggingface.co/ya9919/fin-laya-3-rounds) and Methods M4/M6 give native BF16 and expanded-test results.

## 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. The deployment format is Laya-MLX FP16. `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.