Downloads · 30 days
13
25% of all-time downloads
haofue2i1z3/Qwen3-8B-LegalCitation-GRPO-200
Qwen3-8B-LegalCitation-GRPO-200 is a text generation model from haofue2i1z3. Use it when you need the model to write or continue text. It is set up for peft. The card lists the license as apache-2.0.
基于 Qwen/Qwen3-8B 的 rank-64 GRPO LoRA。训练奖励只使用可确定验证的法条引用信号,覆盖 nocontext、ragclean 和 ragdistractor 三种模式。
Downloads · 30 days
13
25% of all-time downloads
All-time downloads
51
Public
Repo size
710 MB
Likes
0
Public
Click a slice to open those files.
.safetensors698 MB · 98%
From the Hugging Face model README
基于 Qwen/Qwen3-8B 的 rank-64 GRPO LoRA。训练奖励只使用可确定验证的法条引用信号,覆盖 no_context、rag_clean 和 rag_distractor 三种模式。
| 指标 | Base | GRPO-200 |
|---|---|---|
| no-context 无依据引用行 | 40/42 | 36/42 |
| RAG grounded citation precision | 0.7703 | 0.7874 |
| RAG required 行召回率 | 0.8333 | 0.9286 |
| RAG required marker coverage | 0.6765 | 0.7794 |
| 正常停止行 | 41/84 | 49/84 |
| 语义均分变化 | - | -0.039 |
模型最终判定为 FAIL:引用精度、no-context 无依据引用、停止与重复等硬门禁仍未全部通过。它是行为移动研究 checkpoint,不是部署就绪模型。
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
adapter = "haofue2i1z3/Qwen3-8B-LegalCitation-GRPO-200"
base = "Qwen/Qwen3-8B"
tokenizer = AutoTokenizer.from_pretrained(adapter, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(base, device_map="auto", torch_dtype="auto")
model = PeftModel.from_pretrained(model, adapter)