umarfarookm/UmarTransit-1B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

UmarTransit-1B is a 1.54 billion parameter domain-specific language model developed by umarfarookm, fine-tuned from Qwen2.5-1.5B-Instruct. It specializes in public transit systems and GTFS (General Transit Feed Specification) data, excelling at GTFS understanding, transit route analysis, and journey planning. The model is optimized for tasks like validating GTFS feeds and analyzing transit network intelligence.

Loading preview...

UmarTransit-1B: A Specialized LLM for Public Transit

UmarTransit-1B is a 1.54 billion parameter language model, fine-tuned from Qwen/Qwen2.5-1.5B-Instruct, specifically designed for public transit systems and GTFS (General Transit Feed Specification) data. Developed by umarfarookm, this model leverages QLoRA fine-tuning on a dataset of 3,501 instruction pairs from the UmarTransit-Instruct-3k dataset, which is derived exclusively from publicly available GTFS feeds.

Key Capabilities

  • GTFS Understanding & Validation: Proficient in interpreting and validating GTFS data structures and content.
  • Transit Analysis: Capable of analyzing transit routes, schedules, and network statistics.
  • Journey Planning Logic: Improved performance in understanding and processing journey planning queries compared to its base model.
  • Domain-Specific Knowledge: Covers concepts related to transit operations, stop/station information, and transfer logic.

Performance Highlights

UmarTransit-1B (v1.0) demonstrates significant improvements over its base model, achieving an 86% increase in combined evaluation score. It outperforms the base model across all 6 evaluated categories, including GTFS Terminology, GTFS Validation, Route Analysis, Journey Planning, Schedule Reasoning, and Transit Operations. The model has a maximum context length of 1,024 tokens.

Good For

  • Developers building applications requiring deep understanding of public transit data.
  • Analyzing and validating GTFS feeds programmatically.
  • Generating insights into transit network performance and operations.
  • Creating specialized AI assistants for public transit information.

Limitations

It is important to note that UmarTransit-1B is English-only, handles static schedule data only (no real-time predictions), is not a trip planner, and has a limited context of 1,024 tokens.

Finetunes
1 models