WikiofAuto
+28.6% Strict Nested JSON Schema Compliance & Zero-Hallucination MCP Tool CallsMethod: Unsloth QLoRA (r=32) + GRPO Schema Reward Verification on 45k Tool TracesSize: 14.8B

AutoNode-JSON-14B (n8n & MCP Function-Calling Specialist)

Fine-tuned specifically for deterministic n8n AI Agent nodes, Model Context Protocol (MCP) tool invocation, and deeply nested Zod JSON schema extraction without markdown wrapper corruption.

BASE ARCHITECTURE
Qwen/Qwen2.5-14B-Instruct
VRAM REQUIREMENT
10 GB VRAM
QUANTIZATION FORMATS
Q4_K_M (8.9 GB), Q5_K_M (10.3 GB), Q8_0 (15.7 GB)
PARAMETERS
14.8B
QUICK DEPLOY COMMAND (OLLAMA / LLAMA.CPP):ollama run qwen2.5:14b-instruct-q4_K_M

Why Base 8B–14B Models Fail in Production n8n Pipelines

In multi-step n8n workflows, a single malformed JSON key or conversational preamble ("Sure! Here is the JSON...") breaks downstream Switch and HTTP Request nodes. AutoNode-JSON-14B is fine-tuned with Group Relative Policy Optimization (GRPO) using a deterministic JSON-schema validator reward function, achieving 99.7% parseable output on the first pass.