mamadzebal/HieraMind-I2W-Qwen3-4B

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 20, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The mamadzebal/HieraMind-I2W-Qwen3-4B is a 4 billion parameter instruction-tuned causal language model, fine-tuned from Qwen/Qwen3-4B-Thinking-2507. Developed by mamadzebal, this model specializes in generating structured workflow Directed Acyclic Graphs (DAGs) from natural language service requests and retrieved service catalog contexts. It is specifically designed for intent-to-workflow (I2W) tasks within hierarchical service orchestration frameworks, achieving high precision and recall in node generation. With a context length of 32768 tokens, it is optimized for complex workflow generation in technical domains.

Loading preview...

HieraMind-I2W-Qwen3-4B Overview

This model, developed by mamadzebal, is a 4 billion parameter instruction-tuned causal language model, fine-tuned from Qwen/Qwen3-4B-Thinking-2507. It is a core component of the HieraMind framework's Intent-to-Workflow (I2W) system, designed for intent-aware service orchestration.

Key Capabilities

  • Workflow DAG Generation: Given a natural-language service request and a relevant service catalog context, the model generates a structured workflow DAG. This DAG specifies required service functions and their execution dependencies.
  • Reinforcement Learning Fine-tuning: The model was fine-tuned using group-based reinforcement learning, guided by a teacher for workflow evaluation. Semantic retrieval is employed to refine the service catalog provided during workflow generation.
  • High Performance on I2W Tasks: On the HieraMind I2W evaluation set, the model demonstrates strong performance with approximately 96.0% node precision and 92.6% node recall, indicating its effectiveness in accurately identifying and structuring workflow elements.

Good For

  • Automated Service Orchestration: Ideal for systems requiring the automatic translation of user intents into executable service workflows.
  • Complex System Automation: Suitable for environments with heterogeneous network and computing infrastructures where hierarchical control is beneficial.
  • Structured Output Generation: Excels in tasks where the output needs to conform to a specific structured format, such as a DAG, based on natural language input.