Pipeline mind map
Standard pipeline template for a text competition. Not yet specialised by an editor.
Data processing
- Clean and tokenize text
- Check label balance
Model design
- TF-IDF + linear baseline
- Pretrained transformer
Training
- Folds by label
- Small learning rate, few epochs
Post-processing
- Calibrate / threshold for the metric
- Ensemble seeds
Metric
Log Loss. Log loss: punishes confident wrong probabilities; lower is better.
Public baselines
Most-voted public notebooks, refreshed daily from the Kaggle API (last 2026-10-10).
- LMSYS: KerasNLP Starter ▲1606
- finetuning_test2 ▲152
- [HCMUS][2025][24C15034] Ensemble Inference ▲123
- LLM Classification Finetuning | ML LightGBM ▲108
- Inference - llama-3 8b ▲79
- Model's Battles on LLM Finetuning ▲79
- GATS-LLM-Finetuning-Classification ▲63
- [Inference] Gemma-2 9b 4-bit QLoRA 6b251d ▲61
- Inference of Llama and Gemma(final) ▲57
- Inference - llama-3 8b ▲56
- [Training]Gemma-2 9b 4-bit QLoRA fine-tunin 25a55e ▲51
- LLM Classification finetuning ▲50
Discussion 0
No discussion yet. Ask the first question or post a team-up for this competition.
Sign in with PocketPlay
Sign-in opens soon. Until then the forum is read-only.
Same account as pocketplay.win (Google or username). Posting needs a Google-linked account.
Markdown: **bold**, *italic*, `code`, ``` code blocks, - lists, > quotes, [text](https://link). No HTML.