MainStack/marvy-1-14B

TEXT GENERATIONConcurrency Cost:1Model Size:14.8BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 31, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

MainStack/marvy-1-14B is a 14 billion parameter language model, fine-tuned from Qwen2.5-14B-Instruct, specifically designed for the full ServiceNow delivery lifecycle. Developed by MainStack, it excels at drafting implementation-grade artifacts such as business analyses, requirements, solution design documents, and test cases. This model is optimized to produce structured content for ServiceNow practitioners, offering significant perplexity reduction on domain-specific tasks compared to its base model.

Loading preview...

marvy-1-14B: ServiceNow Delivery Lifecycle Specialist

marvy-1-14B is a 14 billion parameter language model developed by MainStack, fine-tuned from Qwen2.5-14B-Instruct. Its core purpose is to generate implementation-grade first drafts for the entire ServiceNow delivery lifecycle, from initial business analysis to final validation. Unlike general-purpose LLMs, marvy is trained on 1,958 anonymized, real-world ServiceNow engagement artifacts (887k tokens), ensuring its outputs align with the structure and sequence of actual delivery processes.

Key Capabilities

  • Comprehensive Artifact Generation: Drafts a wide range of documents including business analysis reports, functional/non-functional requirements, stakeholder mappings, systems inventories, Solution Design Document sections, user stories with acceptance criteria, implementation plans, and test cases.
  • OOTB-first and Implementation-Grade: Tuned to prioritize out-of-the-box ServiceNow correctness, producing drafts that require review and refinement rather than complete rewrites.
  • Performance: Achieves a 32.3% perplexity reduction (from 8.91 to 6.03) on unseen customer data compared to its base model, with significant gains on structured artifacts like systems inventories (86.3% improvement) and requirements extraction (79.9% improvement).
  • Local Deployment: Available in merged FP16, LoRA adapter, and GGUF quants, enabling local and private execution on platforms like Apple Silicon via LM Studio or Ollama.

Good For

  • ServiceNow Practitioners: Ideal for solution architects, business analysts, technical consultants, and project managers seeking to accelerate the creation of delivery artifacts.
  • Drafting Structured Content: Excels at generating precise, formatted documents for ServiceNow projects.
  • Maintaining Data Privacy: Designed for local execution, ensuring sensitive engagement data remains on-premises.

Finetunes
3 models