upwitu/qwen3-4b-sft-all

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The upwitu/qwen3-4b-sft-all model is a 4 billion parameter Qwen3-Instruct model, fine-tuned by upwitu on the multi-task CAR-Bench Winner Dataset. This model integrates safety invariants and execution patterns derived from the top three winning architectures of the IJCAI 2026 CAR-Bench competition, including Read-Before-Write, 4-step Chain of Thought reasoning, and Programmatic CodeAct. It is specifically designed for in-cabin autonomous vehicle AI assistance, excelling in tool-use scenarios with robust policy adherence and structured reasoning.

Loading preview...

Overview

This model, upwitu/qwen3-4b-sft-all, is a 4 billion parameter Qwen3-Instruct model, fine-tuned by upwitu using a multi-task Supervised Fine-Tuning (SFT) approach. It leverages the CAR-Bench Winner Dataset, which synthesizes safety invariants and execution patterns from the top three winning architectures of the IJCAI 2026 CAR-Bench competition. This specialized training makes it particularly adept at complex, safety-critical reasoning and tool-use within autonomous systems.

Key Capabilities

  • Advanced Safety Protocols: Incorporates principles like Read-Before-Write, Confirmation Gates, and Actuator Boundary Gates from the 10CARS architecture.
  • Structured Reasoning: Employs a 4-step Chain of Thought (CoT) process (Context Audit -> Policy Check -> Tool Selection & Provenance -> Execution Plan) for robust decision-making, inspired by FreudeDrive.
  • Programmatic Tool Use: Features Programmatic CodeAct and Policy-as-Code, enabling sophisticated interaction with external systems and tools, derived from Proxima Ultra.
  • Tool Call Parsing: Designed to work seamlessly with a Hermes Tool Call Parser for structured function calls, crucial for in-cabin vehicle systems.

Good For

  • In-Cabin Autonomous Vehicle AI: Its primary use case, providing intelligent assistance with strong safety and policy adherence.
  • Complex Tool-Use Agents: Ideal for applications requiring an LLM to interact with a predefined set of tools in a structured and safe manner.
  • Safety-Critical AI Systems: Suitable for environments where robust policy enforcement and verifiable execution patterns are paramount.
  • Research in AI Safety and Explainability: The integrated CoT and policy-as-code features offer a strong foundation for studying and developing more transparent and controllable AI behaviors.

Finetunes
2 models