TinyModel1

TinyModel1 is a compact encoder model for news topic classification, trained on the AG News dataset. It targets fast CPU/GPU inference and use as a baseline.

Links

• Source code (train & export): https://github.com/HyperlinksSpace/TinyModel
• Live demo (Space): TinyModel1Space (canonical Hub URL; avoids unreliable *.hf.space links)

---

Model summary

Field	Value
Task	Text classification (single-label, 4 classes)
Labels	World, Sports, Business, Sci/Tech
Dataset	fancyzhx/ag_news
Architecture	Tiny BERT-style encoder (BertForSequenceClassification)
Parameters	1,339,268 (~1.34M)
Max sequence length	128 tokens (training & inference)
Framework	Transformers · Safetensors

---

Model overview

Trained with a WordPiece tokenizer fit on the training split and a shallow BERT stack. Replace the dataset and labels via scripts/train_tinymodel1_classifier.py for your own taxonomy.

Core capabilities

• Text routing — assign one class per input for search, feeds, or triage.
• Low latency — small parameter count suits edge and serverless setups.
• Fine-tuning base — swap labels or data for your domain while keeping the same architecture.

---

Training

Setting	Value
Train samples (cap)	3000
Eval samples (cap)	600
Epochs	2
Batch size	16
Learning rate	0.0001
Optimizer	AdamW

---

Evaluation

Metric	Value
Accuracy	0.5383
Macro F1	0.4554
Weighted F1	0.4527
Final train loss	1.1567

Per-class F1 and the confusion matrix are saved in eval_report.json in this model directory.

Metrics are computed on the held-out eval subset (see eval_report.json → reproducibility); treat them as a sanity-check baseline, not a production SLA.

---

Getting started

Inference with transformers

from transformers import pipeline

clf = pipeline(
    "text-classification",
    model="TinyModel1",
    tokenizer="TinyModel1",
    top_k=None,
)
text = "Your input text here."
print(clf(text))

Use top_k=None (or your Transformers version’s equivalent) for scores for all labels. Replace "TinyModel1" with your Hub model id when loading from the Hub.

---

Training data

• Dataset: fancyzhx/ag_news (text column mapped for training; see artifact.json).
• Preprocessing: tokenizer trained on training texts; sequences truncated to 128 tokens.

---

Intended use

• Prototyping routing, tagging, and dashboard features over short text.
• Teaching and benchmarking small-classification setups.
• Starting point for domain adaptation with your own labels.

---

Limitations

• Accuracy is modest by design; validate on your data before high-stakes use.
• Not a general-purpose language model — classification head only; for generation use an LM.
• Tokenizer and labels are tied to this training run; mismatched inputs may degrade.

---

License

This model is released under the Apache 2.0 license (see repository LICENSE where applicable).