Instructions to use zhatrix/logistics-qwen3-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use zhatrix/logistics-qwen3-lora with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-14B") model = PeftModel.from_pretrained(base_model, "zhatrix/logistics-qwen3-lora") - Notebooks
- Google Colab
- Kaggle
物流通 · Logistics LoRA for Qwen3-14B
物流行业智能助手的 LoRA 适配器(PEFT 格式,v2.2)。配合知识库检索与业务工具使用,覆盖客服问答、单据/地址抽取、路径调度、行业知识四类场景。
- 基座:
Qwen/Qwen3-14B(混合推理模型;业务场景建议enable_thinking=False) - LoRA:r=8,alpha=160,作用于后 16 层的 q/k/v/o/gate/up/down
- 训练:mlx_lm LoRA 800 步(≈1 epoch,1472 条样本),Val loss 0.032(验证集与训练集按用户问题分组切分、零泄漏)
- 数据:教师问答人工精编 166 条(含数值比较与分档边界靶向题);地址样本用真实省市区;含工具失败/边界轨迹、运费+时效组合调用、多轮追问样本
- 评测(项目自带 55 例规则评测:数值边界/失败场景/多轮/组合调用):v2.1 0.909 → 本适配器 0.982(客服 1.0 / 知识 0.958 / 抽取 1.0 / 调度 1.0);早期 20 例子集 20/20
上一代(Qwen2.5-7B 基座,0.880):logistics-qwen-lora。
使用
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = "Qwen/Qwen3-14B"
tok = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, "<本仓库ID>")
# 生成时模板传 enable_thinking=False,工具调用为 Qwen hermes JSON 格式
完整应用(检索、10 个业务工具、Gradio/FastAPI):https://github.com/zhatrix/logicLLM
局限
演示用途;运单/网点数据为模拟数据。训练分布外的数值套档偶有错误。
- Downloads last month
- 34
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support