TIME-Module: Mapping — flan-t5-small

Model Description

Temporal span prediction model trained on the combined dataset (quality + difficulty splits merged). Provides broader coverage across difficulty tiers compared to the quality-only model.

Training Details

• Base Model: google/flan-t5-small
• Architecture: T5ForConditionalGeneration
• Dataset: Pieces/temporal-span-prediction-combined-v1 (54,500 train (quality + difficulty combined))
• Training Steps: 50,000
• Learning Rate: 3e-4
• Batch Size: 16 × 2
• Mixed Precision: bf16
• Hardware: NVIDIA RTX 4090 (24 GB)

Results

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained('Pieces/time-mapping-flan-t5-small-combined-best')
model = AutoModelForSeq2SeqLM.from_pretrained('Pieces/time-mapping-flan-t5-small-combined-best')
inputs = tokenizer('Map: tomorrow at 3pm | ref_time: 2026-01-15T10:00:00 | tz: UTC', return_tensors='pt')
outputs = model.generate(**inputs)
result = tokenizer.decode(outputs[0], skip_special_tokens=True)

Part of the TIME-Module Project

This model is part of the TIME (Temporal Intent, Mapping, and Extraction) module, a suite of models for understanding and processing temporal information in natural language.

Related models:

• Pieces/time-classification-flan-t5-small-split-best — Intent classification
• Pieces/time-mapping-flan-t5-small-quality-best — Span prediction (best)
• Pieces/time-mapping-t5gemma-270m-best — Span prediction (T5Gemma)

Citation

@software{time_module,
  title={TIME-Module: Temporal Intent, Mapping, and Extraction},
  author={Pieces},
  year={2026},
  url={https://huggingface.co/Pieces}
}