初始化项目，由ModelHub XC社区提供模型

Model: 84basi/lora-10-1
Source: Original Platform

README.md

@@ -0,0 +1,59 @@
+---
+base_model: unsloth/Qwen3-4B-Instruct-2507
+datasets:
+- u-10bei/structured_data_with_cot_dataset_512_v5
+language:
+- en
+license: apache-2.0
+library_name: transformers
+pipeline_tag: text-generation
+tags:
+- full-finetune
+- structured-output
+---
+
+qwen3-4b-structured-output-lora
+
+This repository provides a **full fine-tuned model** based on
+**unsloth/Qwen3-4B-Instruct-2507** using **BF16 full fine-tuning + NEFTune**.
+
+This repository contains the **complete model weights**.
+
+## Training Objective
+
+This model is trained to improve **structured output accuracy**
+(JSON / YAML / XML / TOML / CSV).
+
+CoT (Chain-of-Thought) is removed from training data during preprocessing.
+
+## Training Configuration
+
+- Base model: unsloth/Qwen3-4B-Instruct-2507
+- Method: Full fine-tuning (BF16)
+- Max sequence length: 2048
+- Epochs: 1
+- Learning rate: 2e-05
+- NEFTune noise alpha: 5.0
+
+## Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+model_id = "84basi/lora-10-1"
+
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+)
+```
+
+## Sources & Terms (IMPORTANT)
+
+Training data: u-10bei/structured_data_with_cot_dataset_512_v5
+
+Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License.
+Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.