ya9919/fin-laya-3-rounds

Fin-Laya — 3-rounds

This is the standalone release card for the 3-rounds checkpoint. Weights and runtime files remain hosted in the original repository at the two-source revision. Use the download command below to retrieve this version.

English financial decision model, approximately 0.42B parameters. This card describes the 3-rounds release (previously labeled two-source). Training data cover CHFS and CGSS. The training sequence is SFT → SFT → native RLCD.

Models collection · 6-rounds model card

Evaluation

Test scope	Questions	Accuracy	Macro F1
CHFS + CGSS	240,458	74.5336%	38.0164%
CHFS + CGSS + CFPS + CLDS	418,597	56.3657%	19.2395%

Macro F1 averages over all candidate classes within each supported question version, then equally over versions.

Inference speed

8,822 input tokens/s, 26.94 ms mean latency, 37.12 decisions/s. Measured on one A100 40GB with FP16, batch and concurrency one, the same three requests, three warmups and 20 timed calls per request; input throughput uses a shared tokenizer.

Download and reproduce

hf download ya9919/fin-laya --revision two-source --local-dir fin-laya-3-rounds
cd fin-laya-3-rounds
python evaluate.py --output evaluation_reproduced.json

Evaluation report · Evaluation script · MAS Simulation

中文

本页为 3-rounds 训练版本，训练顺序为 SFT → SFT → 原生 RLCD，数据来源为 CHFS、CGSS。下载使用保留的 two-source 标签，确保取得本版本权重；6-rounds 使用另一张模型卡。