SEO/GEO: rewrite intro with keyword-rich quotable authoritative content

README.md

@@ -72,12 +72,21 @@ pipeline_tag: text-generation
 
 ---
 
-# MiniCPM5-1B-Agentic-Tooluse-QLoRA-v3
+# MiniCPM5-1B-Agentic-Tooluse-QLoRA-v3 — Small Function-Calling LoRA Adapter (GRPO + QLoRA)
 
-**A small, fast, accurate function-calling / tool-use LoRA adapter for [MiniCPM5-1B](https://huggingface.co/openbmb/MiniCPM5-1B).** Built for developers who need reliable agentic tool calling, JSON/XML function-call generation, and structured output from a **1-billion-parameter model** that runs comfortably on a single consumer GPU, a laptop, a phone, or any edge/on-device deployment — no 7B+ model required.
+**MiniCPM5-1B-Agentic-Tooluse-QLoRA-v3** is a **LoRA adapter for MiniCPM5-1B** that turns a 1-billion-parameter base model into a reliable function-calling / tool-calling agent. Under 100 MB download. Load it with PEFT on top of [openbmb/MiniCPM5-1B](https://huggingface.co/openbmb/MiniCPM5-1B) and run it on a consumer GPU, a laptop, or any edge device.
 
-If you're searching for a **small LLM for function calling**, a **local tool-calling model for llama.cpp / Ollama / vLLM**, an **on-device agent backbone**, or a **cheap, fast alternative to GPT-4o / Claude function calling for structured agent tasks**, this adapter is built exactly for that use case.
+If you are looking for a **small LLM for function calling**, a **lightweight tool-use LoRA adapter**, a **local AI agent backbone under 100MB**, a **cheap fast alternative to GPT-4o / Claude function calling**, or a **GRPO-trained structured-output model**, this adapter is built exactly for that.
 
+> **74.67% exact-argument accuracy** on a held-out 300-example benchmark — trained with QLoRA supervised fine-tuning followed by GRPO reinforcement learning, rewarding exact function-name and argument-value correctness. The adapter itself is under 100 MB.
+
+## Why MiniCPM5-1B-Agentic-Tooluse?
+
+- **Tiny footprint, real accuracy.** 1B parameters, adapter under 100 MB — deployable anywhere a 7B+ model can't go: mobile apps, browser extensions, IoT/embedded agents, offline assistants, cost-sensitive high-throughput API backends.
+- **Purpose-built for agentic tool use.** Trained specifically to parse a tool/function schema plus a natural-language user request and emit a correctly-named, correctly-structured, correctly-valued function call — the core skill every LLM agent framework (LangChain, LlamaIndex, AutoGen, CrewAI, custom ReAct loops, MCP servers) depends on.
+- **Two-stage training: QLoRA SFT + GRPO reinforcement learning.** Most open tool-calling fine-tunes stop at supervised fine-tuning. This adapter adds GRPO (Group Relative Policy Optimization) RL on top, specifically rewarding exact function-name selection and exact argument-value correctness — the two hardest, most failure-prone parts of tool calling for small models.
+- **Honestly measured, not marketing numbers.** Every metric comes from one evaluation harness run end-to-end on a locked, held-out 300-example test split — same parser, same grader, same slice for the base model, the SFT model, and this GRPO-refined v3 adapter.
+- **Compared to GPT-4o / Claude for function calling:** 100% free, fully local, zero per-call cost, fine-tunable, data never leaves your machine.
 ## Why MiniCPM5-1B-Agentic-Tooluse?
 
 - **Tiny footprint, real accuracy.** 1B parameters total, LoRA adapter itself is under 100MB — deployable anywhere a 7B+ model can't go: mobile apps, browser extensions, IoT/embedded agents, offline assistants, cost-sensitive high-throughput API backends.