Duplicate from Freakz3z/Qwen-JSON

Co-authored-by: HU RUIKE <Freakz3z@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,36 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+Qwen-JSON_Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text

Qwen-JSON_Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ec43272bc26ffe0405ef4f7b132ac6d77809666755e57fd0b35d004287374fd
+size 2497280224

README.md

@@ -0,0 +1,57 @@
+---
+base_model: Qwen/Qwen3-4B-Instruct-2507
+language:
+- en
+library_name: transformers
+license: apache-2.0
+model_name: qwen-json
+pipeline_tag: text-generation
+tags:
+- unsloth
+- trl
+- grpo
+- reinforcement-learning
+- json
+- recipe
+---
+
+# RL-Struct: Bridging the Structure Gap
+
+[中文版本](./README_CN.md) | [📚 Paper](https://huggingface.co/papers/2512.00319)
+
+We introduce **RL-Struct**, a lightweight Reinforcement Learning framework designed to solve the "Structure Gap"—the tension between probabilistic token generation and deterministic structured formats (e.g., JSON). By leveraging **GRPO (Gradient Regularized Policy Optimization)** and a **Multi-dimensional Reward Function**, our model achieves superior structural reliability without the high inference latency of constrained decoding.
+
+## 🚀 Key Features
+
+-   **Multi-dimensional Reward Function**: Decomposes the objective into Structure, Format, Validity, Correctness, and Length.
+-   **Efficient Training**: Uses GRPO to eliminate the critic network, reducing VRAM usage by ~40% compared to PPO.
+-   **Emergent Curriculum**: The model spontaneously learns syntax (how to speak) before semantics (what to say).
+-   **High Performance**: Achieves **89.7% Structural Accuracy** and **92.1% JSON Validity** on complex recipe generation, outperforming LLaMA-3-8B and GPT-3.5.
+
+## 📊 Model Details
+
+-   **Base Model:** [Qwen/Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)
+-   **Training Method:** GRPO (Reinforcement Learning) + LoRA
+-   **Task:** Structured Output Generation (JSON Recipes, GSM8K-JSON, ToolUse)
+-   **License:** Apache-2.0
+
+## 🛠️ Usage
+
+The following is the system prompt:
+
+```text
+You are a precise recipe assistant. Always respond in the following JSON format:
+{
+  "reasoning": "Your step-by-step reasoning here...",
+  "answer": "{\"name\": \"Recipe Name\", \"nutrition\": \"Calories: ..., Protein: ..., Fat: ...\"}"
+}
+Do not include any other text, explanations, or markdown. Only output valid JSON.
+```
+
+## 📈 Performance
+
+| Method | Structural Acc. | JSON Validity | Content Acc. |
+| :--- | :---: | :---: | :---: |
+| GPT-3.5 (Zero-shot) | 45.5% | 82.1% | 88.0% |
+| LLaMA-3-8B (SFT) | 78.2% | 85.4% | 86.0% |
+| **RL-Struct (Ours)** | **89.7%** | **92.1%** | **84.5%** |

README_CN.md

@@ -0,0 +1,43 @@
+# RL-Struct: 弥合结构鸿沟
+
+[English Version](./README.md)
+
+本仓库包含论文 **"Bridging the Structure Gap: A Lightweight RL Framework for Reliable Structured Output Generation in LLMs"** 的模型和代码。
+
+我们提出了 **RL-Struct**，这是一个轻量级的强化学习框架，旨在解决“结构鸿沟”问题——即概率性 Token 生成与确定性结构化格式（如 JSON）之间的矛盾。通过利用 **GRPO（梯度正则化策略优化）** 和新颖的 **多维奖励函数**，我们的模型在无需高延迟约束解码的情况下，实现了卓越的结构可靠性。
+
+## 🚀 核心特性
+
+-   **多维奖励函数**：将目标分解为结构（Structure）、格式（Format）、有效性（Validity）、正确性（Correctness）和长度（Length）。
+-   **高效训练**：使用 GRPO 消除 Critic 网络，相比 PPO 减少约 40% 的显存占用。
+-   **涌现课程学习**：模型自发地先学习语法（如何说），再学习语义（说什么）。
+-   **高性能**：在复杂的食谱生成任务上实现了 **89.7% 的结构准确率** 和 **92.1% 的 JSON 有效性**，优于 LLaMA-3-8B 和 GPT-3.5。
+
+## 📊 模型详情
+
+-   **基座模型：** [Qwen/Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)
+-   **训练方法：** GRPO (强化学习) + LoRA
+-   **任务：** 结构化输出生成 (JSON 食谱, GSM8K-JSON, ToolUse)
+-   **协议：** Apache-2.0
+
+## 🛠️ 使用方法
+
+### 系统提示词 (System Prompt)
+为确保正确的 JSON 输出，请使用以下系统提示词：
+
+```text
+You are a precise recipe assistant. Always respond in the following JSON format:
+{
+  "reasoning": "Your step-by-step reasoning here...",
+  "answer": "{\"name\": \"Recipe Name\", \"nutrition\": \"Calories: ..., Protein: ..., Fat: ...\"}"
+}
+Do not include any other text, explanations, or markdown. Only output valid JSON.
+```
+
+## 📈 性能表现
+
+| 方法 | 结构准确率 | JSON 有效性 | 内容准确率 |
+| :--- | :---: | :---: | :---: |
+| GPT-3.5 (Zero-shot) | 45.5% | 82.1% | 88.0% |
+| LLaMA-3-8B (SFT) | 78.2% | 85.4% | 86.0% |
+| **RL-Struct (Ours)** | **89.7%** | **92.1%** | **84.5%** |