kamaboko2007/llm_advance_024_enhanced_rules
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 27, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The kamaboko2007/llm_advance_024_enhanced_rules model is a 4 billion parameter Qwen3-4B-Instruct-2507 based model, fine-tuned for enhanced performance on AgentBench tasks like ALFWorld and DBBench. It uniquely addresses catastrophic forgetting and format-collision in multi-task agent fine-tuning through a custom Jinja2 chat_template. This model excels at dynamic, context-aware injection of task-specific heuristics and error recovery rules, making it highly effective for agentic workflows.

Loading preview...