Nori-30M is the ~29.2M-parameter variant of Nori, a tabular foundation model for regression via in-context learning (ICL). Given a few labeled rows as context, it predicts on new query rows in a single forward pass, with no task-specific training or fine-tuning. The model is trained entirely on synthetic data. Mean and median R² across 96 regression tasks from three public benchmark suites, on the same protocol as the base Nori: Stronger than the ~6M base on every suite. Evaluated with the bundled default inference config and the large-GPU protocol (up to 50k context rows per dataset). Paste this into Claude Code, Cursor, or any AI coding assistant and it will wire python from synthefynori…
Open-weight model · Tabular classification
transfer-fraud-model
by Youssef Mlaouhia yssfmlha/transfer-fraud-model
transfer-fraud-model is an open-weight model for tabular classification from Youssef Mlaouhia, released under MIT License. Its published files total 712.7 KB.
Gradient-boosted classifier. You compute the 15 features yourself and pass them in; the model returns P(fraud) and a risk level 0-3. Trained on a synthetic dataset - validate on real data.
Parameters—
Context—
Weights712.7 KB
Licensemit
AccessOpen weights
Monthly Downloads—
Model Card
By Youssef Mlaouhia, published under mit, revision 273172cbaeb4.
Gradient-boosted classifier. You compute the 15 features yourself and pass them in; the model returns P(fraud) and a risk level 0-3. Trained on a synthetic dataset - validate on real data. predict accepts one dict, a list of dicts, or a pandas DataFrame with these columns (extra columns are ignored, order doesn't matter, missing ones raise an error). "History" = the sender's transfers strictly before the one being scored, within the last 180 days. "rocauctest": 0.9924, "prauctest": 0.8382, "fraudratetest": 0.0169, "alertprecisionlevel2plus": 0.64, "alertrecalllevel2plus": 0.8287, "recalllevel1plus": 0.9655
Read Youssef Mlaouhia's full model card
Bank transfer fraud model (feature input)
Gradient-boosted classifier. You compute the 15 features yourself and pass them in; the model returns P(fraud) and a risk level 0-3. Trained on a synthetic dataset - validate on real data.
Usage
from huggingface_hub import snapshot_download
import sys
folder = snapshot_download("YOUR_USERNAME/YOUR_REPO"); sys.path.insert(0, folder)
from predict import FraudModel
model = FraudModel.load(folder)
print(model.predict({'amount': 361.71, 'hour': 2.115, 'day_of_week': 2.0, 'history_count': 87.0, 'is_new_destination': 0.0, 'times_sent_to_destination': 3.0, 'days_since_last_to_destination': 0.0373, 'amount_vs_avg': 1.3774, 'amount_zscore': 0.3393, 'amount_vs_max': 0.2473, 'hour_familiarity': 0.046, 'hours_since_last_transfer': 0.6567, 'transfers_last_1h': 4.0, 'transfers_last_24h': 4.0, 'distinct_destinations_last_24h': 2.0}))
predict accepts one dict, a list of dicts, or a pandas DataFrame with these columns (extra columns are ignored,
order doesn't matter, missing ones raise an error).
Input features
| feature | definition |
|---|---|
amount |
Amount of the transfer being checked (same currency as in training). |
hour |
Hour of day as a decimal in the customer's LOCAL time (14:30 -> 14.5), range 0-24. |
day_of_week |
Monday=0 ... Sunday=6. |
history_count |
Number of the sender's earlier transfers in the last 180 days. |
is_new_destination |
1 if the sender never sent to this destination account in the last 180 days, else 0. |
times_sent_to_destination |
How many earlier transfers (last 180 days) went to this destination. |
days_since_last_to_destination |
Days since the sender's last transfer to this destination; -1 if never. |
amount_vs_avg |
amount / average amount of the sender's earlier transfers; 1.0 if no history. |
amount_zscore |
(amount - mean of earlier amounts) / std of earlier amounts; 0.0 if no history or std = 0. |
amount_vs_max |
amount / largest earlier transfer amount of the sender; 1.0 if no history. |
hour_familiarity |
Share (0-1) of the sender's earlier transfers made within +-2h of this hour (circular: 23:00 and 01:00 are 2h apart); 0.5 if no history. |
hours_since_last_transfer |
Hours since the sender's previous transfer (any destination); -1 if no history. |
transfers_last_1h |
Number of the sender's earlier transfers in the previous 1 hour. |
transfers_last_24h |
Number of the sender's earlier transfers in the previous 24 hours. |
distinct_destinations_last_24h |
Number of different destination accounts among those last-24h transfers. |
"History" = the sender's transfers strictly before the one being scored, within the last 180 days.
Risk levels (probability cut-offs: [0.0053, 0.0819, 0.5633])
| level | label | test transfers | fraud | fraud share |
|---|---|---|---|---|
| 0 | low | 41,273 | 26 | 0.1% |
| 1 | elevated | 2,355 | 103 | 4.4% |
| 2 | high | 473 | 168 | 35.5% |
| 3 | critical | 502 | 456 | 90.8% |
Test metrics
{ "roc_auc_test": 0.9924, "pr_auc_test": 0.8382, "fraud_rate_test": 0.0169, "alert_precision_level2plus": 0.64, "alert_recall_level2plus": 0.8287, "recall_level1plus": 0.9655 }
Identity and Version
- Repository
- yssfmlha/transfer-fraud-model
- Publisher
- Youssef Mlaouhia
- Task
- Tabular classification
- Modality
- Tabular
- Library
- sklearn
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 273172cbaeb4080112124721ac213b026c86a963
- First published
- 2026-10-16
- Last updated
- 2026-10-16
Files and Weights
5 files, 712.7 KB in total.
Configuration2 files · 4.8 KB
Documentation1 file · 3.4 KB
Other1 file · 703.0 KB
Repository1 file · 1.5 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| config.json | Configuration | 2.9 KB | — |
| predict.py | Configuration | 1.9 KB | — |
| README.md | Documentation | 3.4 KB | — |
| model.joblib | Other | 703.0 KB | 8cba4b474792 |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- mit
- Access
- Open weights, no gate
Download from Youssef Mlaouhia
Released by Youssef Mlaouhia through its official repository on Hugging Face. Read the license.
Questions About transfer-fraud-model
Can I use transfer-fraud-model commercially?
Yes. transfer-fraud-model is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.
Similar Models
Go to SAP-RPT Playground ↗ Note: This model and repository were formerly known as ConTextTab. While the code and repository have now been updated in line with the new name sap-rpt-1-oss, the model checkpoint and functionality remain identical. Implementation of the deep learning model with the inference pipeline described in the paper "ConTextTab: A Semantics-Aware Tabular In-Context Learner". Tabular in-context learning (ICL) has recently achieved state-of-the-art (SOTA) performance on several tabular prediction tasks. Previously restricted to classification problems on small tables, recent advances such as TabPFN and TabICL have extended its use to larger datasets. While being…
Nori is a tabular foundation model for regression via in-context learning (ICL). Given a few labeled rows as context, it predicts on new query rows in a single forward pass, with no task-specific training or fine-tuning. The model is trained entirely on synthetic data. Mean and median R² of the base model across 96 regression tasks from three public benchmark suites (single H200, up to 50K context rows per dataset): Large-N / long-context tables (common in TabArena) are the current focus of the large-table training stages. These numbers are reproducible end-to-end with one command — see Reproducing these numbers. Paste this into Claude Code, Cursor, or any AI coding assistant and it will…
EXAONE Tabular is a transformer-based foundation model for tabular data that solves classification and regression through in-context learning: you pass the labeled rows to fit and the model predicts new rows in a single forward pass — no gradient updates and no per-dataset training. This repository is the exaonetabular inference runtime — a self-contained package that loads a released checkpoint and serves predictions through a small, scikit-learn-style API. The code here is permissively licensed; the released weights are non-commercial — see Both checkpoints are released: EXAONETabularClassifier and EXAONETabularRegressor each fetch their own weights with a single frompretrained() call.…
TabPFN is a transformer-based foundation model for tabular data that leverages prior-data based learning to achieve strong performance on small tabular datasets without requiring task-specific training. For detailed usage examples and best practices, check out: - Python ≥ 3.9 - PyTorch ≥ 2.1 - scikit-learn ≥ 1.0 This repository hosts the production TabPFN-v2 base checkpoints. Files matching the pattern tabpfn-v2-classifier-finetuned-.ckpt are content-identical aliases of the corresponding base checkpoints (e.g. tabpfn-v2-classifier-finetuned-gn2p4bpt-xp6f0iqb.ckpt is identical to tabpfn-v2-classifier-gn2p4bpt.ckpt; tabpfn-v2-classifier-finetuned-zk73skhh.ckpt is identical to…
TabFM is a zero-shot tabular foundation model from Google Research. It supports classification and regression on structured/tabular data with mixed numerical and categorical columns, requiring no fine-tuning or hyperparameter search - training examples are passed as context and predictions are made in a single forward pass. This repository contains the PyTorch weights. For the JAX/Flax weights see You can also load directly using the HuggingFace Hub API: Developed by the Google Research team. - Tabular data with numerical and/or categorical columns - Binary and multiclass classification (up to 10 classes) TabFM uses alternating row and column attention to capture both feature interactions…