Repository navigation
Dataset and Labels
Sudhanshu edited this page Nov 11, 2026
·
1 revision
The entire dataset is embedded as RAW_DATA in the Python script: 30 hand-written customer support messages. Each record is a (text, label) tuple.
| Label name | Class ID |
|---|---|
billing |
0 |
technical |
1 |
account |
2 |
feature_request |
3 |
LABEL2ID and ID2LABEL are derived from LABEL_NAMES, and are passed to AutoModelForSequenceClassification for readable predictions.
The stratified train_test_split uses test_size=0.2 and random_state=42 to create 24 train / 6 validation samples. A 6-sample validation split is far too small to estimate production performance reliably.
flowchart TD
D["RAW_DATA: (text, label) tuples"] --> V["LABEL2ID: 4 classes"]
V --> S["train_test_split: 20% validation, seed=42, stratified"]
S --> T["train: 24 records"]
S --> U["validation: 6 records"]
T --> DT["DatasetDict(train)"]
U --> DV["DatasetDict(validation)"]
DT --> Z["tokenizer: truncation, max_length=128"]
DV --> Z
Z --> C["DataCollatorWithPadding per batch"]
C --> F["input_ids, attention_mask, label"]
The sample text is synthetic/hand-written within the uploaded project. Do not replace it with real customer conversations without privacy and retention review.
Getting started
Architecture
- System Context
- High-Level Design
- Low-Level Design
- Model Components
- Training Sequence
- Data & Labels
- Tokenization
Training & results
Engineering reference