The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

SakThai Kaggle Notebooks & Deploy Scripts

Exact training notebooks, job scripts, and deploy scripts used to build the SakThai family

Overview

This dataset contains the exact notebooks and scripts used to train, evaluate, and deploy the SakThai model family — all built on free Kaggle T4 GPUs.

Not a tabular dataset. This repo stores code and notebooks in the Hub's dataset namespace so the whole pipeline ships in one place. There are no data files, so load_dataset() is not applicable — use the Hugging Face Hub file APIs shown in Loading.

Contents: 29 files · 243.5 KB of code & notebooks (excl. this README) License: MIT

Dataset Statistics

Metric	Value
Files	29
Total size (code files)	243.5 KB
Training notebooks	2
Training scripts	5
Job / experiment scripts	6
Evaluation scripts	5
Data prep & augmentation	6
Deploy scripts	3
Infrastructure	2
Downloads	184
Last updated	2026-07-31

File Inventory

All 29 files with live sizes:

Training notebooks

File	Size
`sakthai-7b-engine.ipynb`	24.0 KB
`sakthai-engine.ipynb`	10.5 KB

Training scripts

File	Size
`train-sakthai-0.5b-v2.py`	6.8 KB
`scripts/sakthai-7b-post-train.py`	13.7 KB
`scripts/train-sakthai-1.5b-v2.py`	4.6 KB
`scripts/train-sakthai-coder-browser.py`	4.1 KB
`scripts/train-sakthai-cpu.py`	3.0 KB

Job / experiment scripts (0.5B era)

File	Size
`job-0.5b-exp.py`	12.9 KB
`job-0.5b-hfjobs.py`	7.8 KB
`job-0.5b-nanguard.py`	13.2 KB
`job-0.5b-nanhunt.py`	11.8 KB
`job-0.5b-v7.py`	10.2 KB
`scripts/web-agent-job.py`	5.4 KB

Evaluation scripts

File	Size
`eval-bfcl-0.5b.py`	3.9 KB
`validate.py`	4.9 KB
`validate_exp.py`	3.7 KB
`scripts/eval-light.py`	1.4 KB
`scripts/eval_sakthai_15b_v2_fixed.py`	7.7 KB

Data prep & augmentation

File	Size
`scripts/audit-and-fix-safety-quality.py`	15.7 KB
`scripts/augment-benchmark-targeted.py`	17.5 KB
`scripts/augment-fill-gaps.py`	19.5 KB
`scripts/create-balanced-benchmark.py`	5.2 KB
`scripts/create-benchmark-from-data.py`	6.7 KB
`scripts/generate-browser-data.py`	18.4 KB

Deploy & infra

File	Size
`deploy-endpoint.py`	3.1 KB
`scripts/convert-to-gguf.py`	3.3 KB
`scripts/push-all-to-hub.py`	2.1 KB

Infrastructure

File	Size
`.gitattributes`	2.4 KB
`README.md`	5.7 KB

Category Breakdown

2 training notebooks (sakthai-engine.ipynb, sakthai-7b-engine.ipynb)
5 training scripts (0.5B, 1.5B v2, 7B post-train, coder-browser, CPU)
6 job / experiment scripts (0.5B variants: exp, hfjobs, nanguard, nanhunt, v7)
5 evaluation scripts (BFCL, light eval, 1.5B v2 fixed, validate)
6 data prep & augmentation scripts (audit, gap-fill, benchmark builders, browser data)
3 deploy & infra scripts (endpoint deploy, GGUF conversion, hub push)
2 infrastructure files (.gitattributes, README.md)

Intended Use

Reproducing SakThai model training runs
Understanding the Kaggle/Colab training pipeline
Adapting scripts for your own fine-tuning projects

Methodology

Scripts evolved from v6 era (1.5B and 7B on T4 GPUs) to v7 era (0.5B with lessons from earlier failures). Key v7 improvements: no mid-run Hub pushes, no Trackio logging, bench-exclusion filtering, and prompt masking.

Loading

This repo is code, not tabular data — use the Hub APIs to list and download files:

from huggingface_hub import HfApi, hf_hub_download

# 1. List all files in the repo
files = HfApi().list_repo_files("Nanthasit/sakthai-kaggle-notebooks", repo_type="dataset")
print(len(files))  # 29

# 2. Download a specific script
path = hf_hub_download(
    repo_id="Nanthasit/sakthai-kaggle-notebooks",
    filename="scripts/train-sakthai-1.5b-v2.py",
    repo_type="dataset",
)
print(path)  # local cache path of the downloaded script