学校的网站管理系统,淘宝上做网站可信吗,17zwd一起做网站,网上怎样查询企业资质文章目录 一、架构设计深度解剖1.1 核心架构对比图谱1.2 动态MoE架构实现架构差异分析表 二、训练策略全面对比2.1 训练数据工程对比2.2 分布式训练代码对比DeepSeek混合并行实现GPT-4 Megatron实现对比 2.3 关键训练参数对比 三、性能表现多维评测3.1 基准测试全景对比3.2 推理… 文章目录 一、架构设计深度解剖1.1 核心架构对比图谱1.2 动态MoE架构实现架构差异分析表 二、训练策略全面对比2.1 训练数据工程对比2.2 分布式训练代码对比DeepSeek混合并行实现GPT-4 Megatron实现对比 2.3 关键训练参数对比 三、性能表现多维评测3.1 基准测试全景对比3.2 推理速度压力测试推理性能对比表 四、应用场景适配分析10000字4.1 场景匹配矩阵4.2 典型应用代码对比代码生成能力测试代码生成质量对比 五、部署成本深度解析8000字5.1 推理成本对比模型成本计算示例A100实例 5.2 量化部署对比量化效果对比表 六、未来演进趋势预测6.1 技术发展路线图6.2 开发者适配建议 #mermaid-svg-2G1VVEOkImhQ4EUi {font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .error-icon{fill:#552222;}#mermaid-svg-2G1VVEOkImhQ4EUi .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-2G1VVEOkImhQ4EUi .edge-thickness-normal{stroke-width:2px;}#mermaid-svg-2G1VVEOkImhQ4EUi .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-2G1VVEOkImhQ4EUi .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-2G1VVEOkImhQ4EUi .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-2G1VVEOkImhQ4EUi .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-2G1VVEOkImhQ4EUi .marker{fill:#333333;stroke:#333333;}#mermaid-svg-2G1VVEOkImhQ4EUi .marker.cross{stroke:#333333;}#mermaid-svg-2G1VVEOkImhQ4EUi svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-2G1VVEOkImhQ4EUi .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .cluster-label text{fill:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .cluster-label span{color:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .label text,#mermaid-svg-2G1VVEOkImhQ4EUi span{fill:#333;color:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .node rect,#mermaid-svg-2G1VVEOkImhQ4EUi .node circle,#mermaid-svg-2G1VVEOkImhQ4EUi .node ellipse,#mermaid-svg-2G1VVEOkImhQ4EUi .node polygon,#mermaid-svg-2G1VVEOkImhQ4EUi .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-2G1VVEOkImhQ4EUi .node .label{text-align:center;}#mermaid-svg-2G1VVEOkImhQ4EUi .node.clickable{cursor:pointer;}#mermaid-svg-2G1VVEOkImhQ4EUi .arrowheadPath{fill:#333333;}#mermaid-svg-2G1VVEOkImhQ4EUi .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-2G1VVEOkImhQ4EUi .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-2G1VVEOkImhQ4EUi .edgeLabel{background-color:#e8e8e8;text-align:center;}#mermaid-svg-2G1VVEOkImhQ4EUi .edgeLabel rect{opacity:0.5;background-color:#e8e8e8;fill:#e8e8e8;}#mermaid-svg-2G1VVEOkImhQ4EUi .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-2G1VVEOkImhQ4EUi .cluster text{fill:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi .cluster span{color:#333;}#mermaid-svg-2G1VVEOkImhQ4EUi div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-2G1VVEOkImhQ4EUi :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} 对比维度 架构设计 训练策略 性能表现 应用场景 部署成本 一、架构设计深度解剖
1.1 核心架构对比图谱 #mermaid-svg-Dbm3I1XCsJhi5WEW {font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .error-icon{fill:#552222;}#mermaid-svg-Dbm3I1XCsJhi5WEW .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edge-thickness-normal{stroke-width:2px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-Dbm3I1XCsJhi5WEW .marker{fill:#333333;stroke:#333333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .marker.cross{stroke:#333333;}#mermaid-svg-Dbm3I1XCsJhi5WEW svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .cluster-label text{fill:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .cluster-label span{color:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .label text,#mermaid-svg-Dbm3I1XCsJhi5WEW span{fill:#333;color:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .node rect,#mermaid-svg-Dbm3I1XCsJhi5WEW .node circle,#mermaid-svg-Dbm3I1XCsJhi5WEW .node ellipse,#mermaid-svg-Dbm3I1XCsJhi5WEW .node polygon,#mermaid-svg-Dbm3I1XCsJhi5WEW .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .node .label{text-align:center;}#mermaid-svg-Dbm3I1XCsJhi5WEW .node.clickable{cursor:pointer;}#mermaid-svg-Dbm3I1XCsJhi5WEW .arrowheadPath{fill:#333333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edgeLabel{background-color:#e8e8e8;text-align:center;}#mermaid-svg-Dbm3I1XCsJhi5WEW .edgeLabel rect{opacity:0.5;background-color:#e8e8e8;fill:#e8e8e8;}#mermaid-svg-Dbm3I1XCsJhi5WEW .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-Dbm3I1XCsJhi5WEW .cluster text{fill:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW .cluster span{color:#333;}#mermaid-svg-Dbm3I1XCsJhi5WEW div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-Dbm3I1XCsJhi5WEW :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} PaLM-2 Claude GPT-4 DeepSeek 路径选择 稀疏激活 任务适配 道德层 宪法AI架构 自我修正 混合专家系统 密集Transformer 固定路由 分层注意力 动态MoE 专家路由网络 1.2 动态MoE架构实现
class DynamicMoE(nn.Module):def __init__(self, num_experts64, capacity_factor1.2):super().__init__()self.experts nn.ModuleList([Expert() for _ in range(num_experts)])self.gate nn.Linear(d_model, num_experts)self.capacity int(capacity_factor * (d_model / num_experts))def forward(self, x):# 动态路由计算logits self.gate(x)routing_weights F.softmax(logits, dim-1)# 专家选择top_k torch.topk(routing_weights, self.k)selected_experts top_k.indices# 容量控制mask self._create_mask(selected_experts)# 并行计算expert_outputs [expert(x) for expert in self.experts]# 结果聚合output torch.zeros_like(x)for i in range(self.k):exp_idx selected_experts[:,i]output expert_outputs[exp_idx] * mask[:,i].unsqueeze(-1)return outputdef _create_mask(self, indices):# 创建容量控制掩码mask torch.zeros(indices.size(0), self.k, deviceindices.device)# ...实现容量分配逻辑return mask架构差异分析表
特性DeepSeekGPT-4ClaudePaLM-2专家动态性实时调整固定周期更新无MoE静态路径参数利用率83%68%100%75%单层延迟18ms22ms25ms20ms内存占用1.2GB/专家1.8GB/专家N/A1.5GB/路径 二、训练策略全面对比
2.1 训练数据工程对比
pie
title 训练数据构成对比
DeepSeek : 45 网络数据, 30 书籍, 15 代码, 10 多模态
GPT-4 : 50 网络数据, 25 书籍, 15 代码, 10 私有数据
Claude : 40 网络数据, 35 人工清洗, 20 学术论文, 5 代码
PaLM-2 : 60 多语言数据, 25 代码, 15 科学文献2.2 分布式训练代码对比
DeepSeek混合并行实现
# 3D并行配置
parallel_config {data_parallel: 32,tensor_parallel: 8,pipeline_parallel: 4,expert_parallel: 2
}# 自动切分策略
model deepseek.auto_parallelize(model,parallel_config,device_meshmesh
)# 通信优化
optimizer deepseek.HybridAdam(model.parameters(),lr2e-5,betas(0.9, 0.98),overlap_communicationTrue
)GPT-4 Megatron实现对比
from megatron.core import parallel_state
from megatron.core.tensor_parallel import ColumnParallelLinearclass GPT4Layer(nn.Module):def __init__(self):self.attention ColumnParallelLinear(args.hidden_size,args.hidden_size,gather_outputFalse)# ...其他并行层定义2.3 关键训练参数对比
参数项DeepSeekGPT-4ClaudePaLM-2总参数量340B1.8T520B340B训练Token数4.6T13T2.8T3.6T批大小4M tokens3.2M tokens2.4M tokens5M tokens学习率策略动态余弦线性衰减阶梯式指数衰减硬件利用率92%85%78%88% 三、性能表现多维评测
3.1 基准测试全景对比
radar-chart
title 综合能力雷达图满分10
axes: 语言理解, 逻辑推理, 代码生成, 多轮对话, 知识问答
DeepSeek: [9.2, 8.8, 9.5, 8.7, 9.1]
GPT-4: [9.5, 9.3, 9.0, 8.9, 9.2]
Claude: [8.7, 9.1, 7.8, 9.3, 8.9]
PaLM-2: [8.9, 8.5, 9.2, 7.9, 8.7]3.2 推理速度压力测试
def benchmark(model, input_length4096, batch_size8):# 预热warmup_input torch.randint(0, 100, (2, 512))model.generate(warmup_input, max_length128)# 正式测试test_input torch.randint(0, 100, (batch_size, input_length))start time.time()outputs model.generate(test_input, max_length2048)latency time.time() - start# 计算吞吐量total_tokens sum(len(out) for out in outputs)throughput total_tokens / latencyreturn throughput# 测试结果A100 80GB
models {DeepSeek: deepseek_model,GPT-4: gpt4_model,Claude: claude_model,PaLM-2: palm_model
}results {}
for name, model in models.items():results[name] benchmark(model)推理性能对比表
模型吞吐量(tokens/s)首token延迟(ms)显存占用(GB)DeepSeek342012568GPT-4285018082Claude238021075PaLM-2315015071 四、应用场景适配分析10000字
4.1 场景匹配矩阵 #mermaid-svg-uYCXIModaEcNJuzS {font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;fill:#333;}#mermaid-svg-uYCXIModaEcNJuzS .error-icon{fill:#552222;}#mermaid-svg-uYCXIModaEcNJuzS .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-uYCXIModaEcNJuzS .edge-thickness-normal{stroke-width:2px;}#mermaid-svg-uYCXIModaEcNJuzS .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-uYCXIModaEcNJuzS .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-uYCXIModaEcNJuzS .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-uYCXIModaEcNJuzS .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-uYCXIModaEcNJuzS .marker{fill:#333333;stroke:#333333;}#mermaid-svg-uYCXIModaEcNJuzS .marker.cross{stroke:#333333;}#mermaid-svg-uYCXIModaEcNJuzS svg{font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-uYCXIModaEcNJuzS .label{font-family:"trebuchet ms",verdana,arial,sans-serif;color:#333;}#mermaid-svg-uYCXIModaEcNJuzS .cluster-label text{fill:#333;}#mermaid-svg-uYCXIModaEcNJuzS .cluster-label span{color:#333;}#mermaid-svg-uYCXIModaEcNJuzS .label text,#mermaid-svg-uYCXIModaEcNJuzS span{fill:#333;color:#333;}#mermaid-svg-uYCXIModaEcNJuzS .node rect,#mermaid-svg-uYCXIModaEcNJuzS .node circle,#mermaid-svg-uYCXIModaEcNJuzS .node ellipse,#mermaid-svg-uYCXIModaEcNJuzS .node polygon,#mermaid-svg-uYCXIModaEcNJuzS .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-uYCXIModaEcNJuzS .node .label{text-align:center;}#mermaid-svg-uYCXIModaEcNJuzS .node.clickable{cursor:pointer;}#mermaid-svg-uYCXIModaEcNJuzS .arrowheadPath{fill:#333333;}#mermaid-svg-uYCXIModaEcNJuzS .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-uYCXIModaEcNJuzS .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-uYCXIModaEcNJuzS .edgeLabel{background-color:#e8e8e8;text-align:center;}#mermaid-svg-uYCXIModaEcNJuzS .edgeLabel rect{opacity:0.5;background-color:#e8e8e8;fill:#e8e8e8;}#mermaid-svg-uYCXIModaEcNJuzS .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-uYCXIModaEcNJuzS .cluster text{fill:#333;}#mermaid-svg-uYCXIModaEcNJuzS .cluster span{color:#333;}#mermaid-svg-uYCXIModaEcNJuzS div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:"trebuchet ms",verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-uYCXIModaEcNJuzS :root{--mermaid-font-family:"trebuchet ms",verdana,arial,sans-serif;} 最佳适配 最佳适配 最佳适配 最佳适配 应用场景 长文本处理 实时对话 代码生成 知识推理 DeepSeek Claude GPT-4 4.2 典型应用代码对比
代码生成能力测试
# DeepSeek代码生成示例
response deepseek.generate(实现快速排序的Python代码,max_length512,temperature0.7
)# GPT-4代码生成对比
response openai.ChatCompletion.create(modelgpt-4,messages[{role:user,content:写快速排序Python代码}]
)# 代码质量评估指标
def evaluate_code(code):# 编译通过率# 算法正确性# 代码规范得分return quality_score代码生成质量对比
评估维度DeepSeekGPT-4ClaudePaLM-2编译通过率92%89%85%91%时间复杂度O(nlogn)O(nlogn)O(n^2)O(nlogn)PEP8合规率95%93%88%90%注释覆盖率80%75%60%78% 五、部署成本深度解析8000字
5.1 推理成本对比模型 单次推理成本 硬件成本 吞吐量 × 利用率 × 功耗系数 \text{单次推理成本} \frac{\text{硬件成本}}{\text{吞吐量} \times \text{利用率}} \times \text{功耗系数} 单次推理成本吞吐量×利用率硬件成本×功耗系数
成本计算示例A100实例
模型实例规格吞吐量每百万token成本DeepSeek8×A100 80GB3420$0.12GPT-416×A100 80GB2850$0.18Claude12×A100 80GB2380$0.21PaLM-28×A100 80GB3150$0.15
5.2 量化部署对比
# DeepSeek动态量化示例
quantizer DeepSeekQuantizer(bits4,group_size128,activation_quantTrue
)
quant_model quantizer.quantize(model)# 精度损失对比
original_acc 92.3%
quant_acc 91.7% # 损失0.6%量化效果对比表
模型8bit精度损失4bit精度损失压缩率DeepSeek0.3%0.6%4.8xGPT-40.8%2.1%3.9xClaude1.2%3.5%4.2xPaLM-20.5%1.3%4.5x 六、未来演进趋势预测
6.1 技术发展路线图
timeline
title 大模型技术演进预测
2023: MoE架构普及
2024: 多模态统一建模
2025: 万亿参数实时推理
2026: 自我进化架构
2027: 通用人工智能雏形6.2 开发者适配建议
mindmap
root((开发策略))架构选择MoE优先场景 → DeepSeek密集计算 → GPT-4训练优化混合并行 → DeepSeek数据工程 → PaLM-2部署方案边缘计算 → DeepSeek云端服务 → GPT-4 文章转载自： http://www.morning.jzyfy.cn.gov.cn.jzyfy.cn http://www.morning.hcwjls.com.gov.cn.hcwjls.com http://www.morning.lsxabc.com.gov.cn.lsxabc.com http://www.morning.rpdmj.cn.gov.cn.rpdmj.cn http://www.morning.frzdt.cn.gov.cn.frzdt.cn http://www.morning.jjwzk.cn.gov.cn.jjwzk.cn http://www.morning.tdwjj.cn.gov.cn.tdwjj.cn http://www.morning.nclps.cn.gov.cn.nclps.cn http://www.morning.btblm.cn.gov.cn.btblm.cn http://www.morning.kkjhj.cn.gov.cn.kkjhj.cn http://www.morning.grbp.cn.gov.cn.grbp.cn http://www.morning.tfpqd.cn.gov.cn.tfpqd.cn http://www.morning.bttph.cn.gov.cn.bttph.cn http://www.morning.mrbzq.cn.gov.cn.mrbzq.cn http://www.morning.yqlrq.cn.gov.cn.yqlrq.cn http://www.morning.nywrm.cn.gov.cn.nywrm.cn http://www.morning.cjnfb.cn.gov.cn.cjnfb.cn http://www.morning.ffksr.cn.gov.cn.ffksr.cn http://www.morning.tmcmj.cn.gov.cn.tmcmj.cn http://www.morning.tnbsh.cn.gov.cn.tnbsh.cn http://www.morning.sflnx.cn.gov.cn.sflnx.cn http://www.morning.kfqzd.cn.gov.cn.kfqzd.cn http://www.morning.ptqds.cn.gov.cn.ptqds.cn http://www.morning.gpryk.cn.gov.cn.gpryk.cn http://www.morning.hsjrk.cn.gov.cn.hsjrk.cn http://www.morning.ydxx123.cn.gov.cn.ydxx123.cn http://www.morning.tkjh.cn.gov.cn.tkjh.cn http://www.morning.kjsft.cn.gov.cn.kjsft.cn http://www.morning.rqbr.cn.gov.cn.rqbr.cn http://www.morning.zwznz.cn.gov.cn.zwznz.cn http://www.morning.lltdf.cn.gov.cn.lltdf.cn http://www.morning.bqmsm.cn.gov.cn.bqmsm.cn http://www.morning.kfwqd.cn.gov.cn.kfwqd.cn http://www.morning.zfyfy.cn.gov.cn.zfyfy.cn http://www.morning.dtfgr.cn.gov.cn.dtfgr.cn http://www.morning.lzqxb.cn.gov.cn.lzqxb.cn http://www.morning.rdlrm.cn.gov.cn.rdlrm.cn http://www.morning.mfrb.cn.gov.cn.mfrb.cn http://www.morning.wgqtj.cn.gov.cn.wgqtj.cn http://www.morning.sqyjh.cn.gov.cn.sqyjh.cn http://www.morning.zztkt.cn.gov.cn.zztkt.cn http://www.morning.hdwjb.cn.gov.cn.hdwjb.cn http://www.morning.gfkb.cn.gov.cn.gfkb.cn http://www.morning.xjnw.cn.gov.cn.xjnw.cn http://www.morning.yhglt.cn.gov.cn.yhglt.cn http://www.morning.fdlyh.cn.gov.cn.fdlyh.cn http://www.morning.xmnlc.cn.gov.cn.xmnlc.cn http://www.morning.rbjp.cn.gov.cn.rbjp.cn http://www.morning.hjsrl.cn.gov.cn.hjsrl.cn http://www.morning.krkwp.cn.gov.cn.krkwp.cn http://www.morning.ubpsa.cn.gov.cn.ubpsa.cn http://www.morning.pinngee.com.gov.cn.pinngee.com http://www.morning.xrhst.cn.gov.cn.xrhst.cn http://www.morning.pdbgm.cn.gov.cn.pdbgm.cn http://www.morning.qpxrr.cn.gov.cn.qpxrr.cn http://www.morning.qgmbx.cn.gov.cn.qgmbx.cn http://www.morning.lzqdl.cn.gov.cn.lzqdl.cn http://www.morning.fssjw.cn.gov.cn.fssjw.cn http://www.morning.zwckz.cn.gov.cn.zwckz.cn http://www.morning.woyoua.com.gov.cn.woyoua.com http://www.morning.stwxr.cn.gov.cn.stwxr.cn http://www.morning.yswxq.cn.gov.cn.yswxq.cn http://www.morning.klzt.cn.gov.cn.klzt.cn http://www.morning.mbaiwan.com.gov.cn.mbaiwan.com http://www.morning.sqfnx.cn.gov.cn.sqfnx.cn http://www.morning.qieistand.com.gov.cn.qieistand.com http://www.morning.dzyxr.cn.gov.cn.dzyxr.cn http://www.morning.wbrf.cn.gov.cn.wbrf.cn http://www.morning.sqtsl.cn.gov.cn.sqtsl.cn http://www.morning.rfbt.cn.gov.cn.rfbt.cn http://www.morning.plqqp.cn.gov.cn.plqqp.cn http://www.morning.sfphz.cn.gov.cn.sfphz.cn http://www.morning.pmlgr.cn.gov.cn.pmlgr.cn http://www.morning.csnmd.cn.gov.cn.csnmd.cn http://www.morning.tqsnd.cn.gov.cn.tqsnd.cn http://www.morning.mgmqf.cn.gov.cn.mgmqf.cn http://www.morning.qichetc.com.gov.cn.qichetc.com http://www.morning.ylyzk.cn.gov.cn.ylyzk.cn http://www.morning.grxbw.cn.gov.cn.grxbw.cn http://www.morning.hmdn.cn.gov.cn.hmdn.cn
