coder3101/QwenPaw-Flash-9B-heretic

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 21, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

coder3101/QwenPaw-Flash-9B-heretic is a 9 billion parameter causal language model, fine-tuned from Qwen3.5-9B and subsequently decensored using the Heretic v1.2.0 tool with Arbitrary-Rank Ablation (ARA). This model is specifically optimized for autonomous agent scenarios within the QwenPaw ecosystem, excelling in tool invocation, command execution, memory management, and multi-step planning. It features a native context length of 262,144 tokens and demonstrates significantly reduced refusals compared to its original counterpart, making it suitable for agentic tasks requiring less restrictive content policies.

Loading preview...

Model Overview

coder3101/QwenPaw-Flash-9B-heretic is a 9 billion parameter causal language model, derived from agentscope-ai/QwenPaw-Flash-9B and subsequently decensored using the Heretic v1.2.0 tool with the Arbitrary-Rank Ablation (ARA) method. This process significantly reduces content refusals, with the 'heretic' version showing 12/100 refusals compared to 96/100 in the original model.

Key Capabilities

This model is deeply optimized for the QwenPaw autonomous agent scenario, demonstrating enhanced performance in:

  • Active Memory Management: Autonomously identifies, stores, and retrieves persistent user preferences and task states for logical consistency.
  • Native File Parsing: Optimized for terminal operations, generating precise CLI commands and executing complex file I/O tasks.
  • Efficient Information Search: Enhanced for web-search tool invocation, with precise search intent recognition and multi-step web navigation.
  • Intelligent Guidance: Built-in awareness of the QwenPaw feature map, proactively suggesting functional paths and troubleshooting.

Technical Details

Fine-tuned from Qwen3.5-9B, it shares architectural parameters including a native context length of 262,144 tokens. The model is a Causal Language Model with a Vision Encoder, post-training stage. Benchmarks indicate substantial improvements across QwenPaw-specific task categories, achieving performance comparable to leading flagship models with lower resource requirements.

Good For

  • Autonomous agent development within the QwenPaw ecosystem.
  • Applications requiring robust tool invocation and multi-step planning.
  • Scenarios where reduced content refusal rates are beneficial.
  • Tasks involving active memory management and complex file system orchestration.

Finetunes
1 models