memorisea/Memorisea-4b-v1
Memorisea-4b-v1 by memorisea is a 4 billion parameter agentic foundation model built on Qwen3-4B-Instruct, specialized for multi-turn function calling, strict schema validation, and defensive vulnerability remediation. It excels in generating syntactically compliant code with 93.3% AST validity and handling compound parallel tool calls, outperforming larger models in specific coding tasks. This model is optimized for autonomous agentic decision-making and security-focused software engineering.
Loading preview...
Overview
Memorisea-4b-v1 is a compact 4-billion parameter agentic foundation model developed by memorisea, fine-tuned from Qwen/Qwen3-4B-Instruct-2507. It focuses on structured agentic decision-making, defensive code syntax, and resilient tool dispatching, delivering high operational consistency in producing valid JSON schemas and executing complex tool calls.
Key Capabilities
- Robust AST Syntax Guarantee: Achieves 93.3% Abstract Syntax Tree (AST) syntactic compliance on competitive coding distributions, surpassing larger dedicated coding models.
- Compound & Parallel Tool Calling: Capable of dispatching multi-stage tool calls and complex nested JSON arguments within
<tool_call>boundaries. - Defensive Software Engineering: Fine-tuned on real-world security patches (CWE/CVE remediation) with parameterized queries and strict input sanitization.
- Strict Constraint Following: Excels at zero-chatter, schema-first outputs and structured Markdown layouts, demonstrating 100.0% constraint compliance.
Benchmark Evaluation
Despite its 4B parameter size, Memorisea-4b-v1 demonstrates competitive performance against 7B models. It achieves 93.3% in Coding (AST Syntax Integrity), outperforming Qwen2.5-Coder-7B (80.0%). It also matches leading 7B architectures in instruction adherence (100.0% on Arena-Hard) and shows strong function calling capabilities (46.7% on BFCL).