fin-laya-mlx is an open-weight model from Li, released under Apache License 2.0. It has 421M parameters. At 16-bit it needs about 1 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 10 downloads a month.
English financial decision model, approximately 0.42B parameters, with MLX CUDA support. This card describes the 6-rounds release (previously labeled four-source). Training data cover CHFS, CGSS, CFPS and CLDS.
Runs On
What it takes to serve fin-laya-mlx (421M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 0.8 GB | 1.0 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.4 GB | 0.5 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.2 GB | 0.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 20, 2026.
fin-laya-mlx on every accelerator the SAVRN Index prices, at every precision
Model Card
By Li, published under apache-2.0, revision bac77786e388.
English financial decision model, approximately 0.42B parameters, with MLX CUDA support. This card describes the 6-rounds release (previously labeled four-source). Training data cover CHFS, CGSS, CFPS and CLDS. Macro F1 averages over all candidate classes within each supported question version, then equally over versions. 23,317 input tokens/s, 10.19 ms mean latency, 98.11 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. 本页为 6-rounds 训练版本，旧标签为 four-source；数据来源为 CHFS、CGSS、CFPS、CLDS。3-rounds 训练版本使用独立模型卡，下载链接指向对应历史权重。按训练轮次命名，数据来源单独说明。
Read Li's full model card
Fin-Laya-MLX — 6-rounds
English financial decision model, approximately 0.42B parameters, with MLX CUDA support. This card describes the 6-rounds release (previously labeled four-source). Training data cover CHFS, CGSS, CFPS and CLDS.
Models collection · 3-rounds model card
Evaluation
| Test scope | Questions | Accuracy | Macro F1 |
|---|---|---|---|
| CHFS + CGSS | 240,458 | 74.5544% | 38.3746% |
| CHFS + CGSS + CFPS + CLDS | 418,597 | 67.2726% | 27.8955% |
Macro F1 averages over all candidate classes within each supported question version, then equally over versions.
Inference speed
23,317 input tokens/s, 10.19 ms mean latency, 98.11 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-mlx --revision four-source --local-dir fin-laya-mlx-6-rounds
cd fin-laya-mlx-6-rounds
python evaluate.py --output evaluation_reproduced.json
Evaluation report · Evaluation script · MAS Simulation
中文
本页为 6-rounds 训练版本，旧标签为 four-source；数据来源为 CHFS、CGSS、CFPS、CLDS。3-rounds 训练版本使用独立模型卡，下载链接指向对应历史权重。按训练轮次命名，数据来源单独说明。
Identity and Version
- Repository
- ya9919/fin-laya-mlx
- Publisher
- Li
- Task
- Not stated by the source
- Modality
- Other
- Library
- laya
- Parameters
- 421M parameters
- Languages
- en
- Revision
- bac77786e388db8888719e64cc95395ed997ee0d
- First published
- 2026-10-12
- Last updated
- 2026-10-17
Files and Weights
48 files, 892.7 MB in total. The weights are 1 file totalling 842.6 MB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 842.6 MB | 2f4f14cfa676 |
| encoder/config.json | Configuration | 2.1 KB | — |
| evaluate.py | Configuration | 13.6 KB | — |
| evaluation_manifest.json | Configuration | 1.9 KB | — |
| evaluation_report.json | Configuration | 6.7 MB | — |
| example_request.json | Configuration | 1.1 KB | — |
| mlx_config.json | Configuration | 167 B | — |
| paired_release_report.json | Configuration | 4.2 KB | — |
| predict_example.py | Configuration | 979 B | — |
| release_manifest.json | Configuration | 677 B | — |
| reports/data_export_report.json | Configuration | 1.2 MB | — |
| reports/data_manifest.json | Configuration | 973.3 KB | — |
| reports/data_snapshot.json | Configuration | 925 B | — |
| reports/data_statistics.json | Configuration | 443.9 KB | — |
| reports/encoding_report.json | Configuration | 3.7 KB | — |
| reports/final_report.json | Configuration | 27.4 MB | 5b8af8233dae |
| reports/mlx_renderer_verification.json | Configuration | 127 B | — |
| reports/mlx_test_report.json | Configuration | 9.6 MB | — |
| reports/old_test_regression_report.json | Configuration | 1.8 KB | — |
| reports/speed_comparison.json | Configuration | 5.1 KB | — |
| reports/training_protocol.json | Configuration | 1.6 KB | — |
| reproduction/benchmark_finance.py | Configuration | 6.8 KB | — |
| reproduction/durable.py | Configuration | 2.4 KB | — |
| reproduction/encode_latest.py | Configuration | 6.1 KB | — |
| reproduction/evaluation_tools.py | Configuration | 7.3 KB | — |
| reproduction/experiment_common.py | Configuration | 8.2 KB | — |
| reproduction/mlx_worker.py | Configuration | 5.8 KB | — |
| reproduction/model_manifest.json | Configuration | 297 B | — |
| reproduction/original_base_manifest.json | Configuration | 364 B | — |
| reproduction/preflight_latest.py | Configuration | 3.6 KB | — |
| reproduction/prepare_fixtures.py | Configuration | 3.0 KB | — |
| reproduction/publish_versions.py | Configuration | 10.0 KB | — |
| reproduction/release_stages.py | Configuration | 14.2 KB | — |
| reproduction/run_experiment.py | Configuration | 10.6 KB | — |
| reproduction/training_core.py | Configuration | 5.6 KB | — |
| rl_agent_config.json | Configuration | 4.4 KB | — |
| tokenizer/special_tokens_map.json | Configuration | 694 B | — |
| versions.json | Configuration | 9.3 KB | — |
| LICENSE | Documentation | 10.2 KB | — |
| Model_Card_3_rounds.md | Documentation | 1.7 KB | — |
| Model_Card_EN.md | Documentation | 2.2 KB | — |
| Model_Card_ZH.md | Documentation | 2.1 KB | — |
| NOTICE | Documentation | 1.3 KB | — |
| README.md | Documentation | 1.6 KB | — |
| requirements.txt | Other | 176 B | — |
| .gitattributes | Repository | 1.6 KB | — |
| tokenizer/tokenizer.json | Tokenizer | 3.6 MB | — |
| tokenizer/tokenizer_config.json | Tokenizer | 20.9 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 842.6 MB
Released by Li through its official repository on Hugging Face. Read the license.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 842.6 MB |
| 16-bit | 0.8 GB |
| 8-bit | 0.4 GB |
| 4-bit | 0.2 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About fin-laya-mlx
How much GPU memory does fin-laya-mlx need?
About 1 GB at 16-bit and 0.3 GB at 4-bit: the weights (421M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run fin-laya-mlx on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use fin-laya-mlx commercially?
Yes. fin-laya-mlx is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.