Nvidia免费模型食用指南

Reno · 2024 年7 月 30 日 14:54

NVIDIA 官方自产自销，新注册用户赠送 1k 积分

简单上手

注册账户

访问 NVIDIA官网，然后右上角点击 Login，在弹窗输入邮箱后点击 Next

输入密码并确认密码以及人类身份后点击创建账户

登录邮箱查看新邮件，点击 Verify Email Address 验证

重复步骤1，点击提交

输入用户名，点击 Create NVIDIA Cloud Account

新注册用户赠送 1k 积分，右上角依次点击头像 - Request More，填写企业信息可以获得 5k 积分

获取密钥

点击 Chat

依次点击 Get API Key - Generate Key - Copy Key 复制 Key

OneAPI 配置

对话模型

baseurl: https://integrate.api.nvidia.com
key: 复制得到的Key
model: yi-large, sea-lion-7b-instruct, starcoder2-7b, starcoder2-15b, dbrx-instruct, deepseek-coder-6.7b-instruct, gemma-7b, gemma-2b, gemma-2-9b-it, gemma-2-27b-it, codegemma-1.1-7b, codegemma-7b, recurrentgemma-2b, granite-34b-code-instruct, granite-8b-code-instruct, breeze-7b-instruct, codellama-70b, llama2-70b, llama3-8b, llama3-70b, llama-3.1-8b-instruct, llama-3.1-70b-instruct, llama-3.1-405b-instruct, phi-3-medium-128k-instruct, phi-3-medium-4k-instruct, phi-3-mini-128k-instruct, phi-3-mini-4k-instruct, phi-3-small-128k-instruct, phi-3-small-8k-instruct, codestral-22b-instruct-v0.1, mamba-codestral-7b-v0.1, mistral-7b-instruct, mistral-7b-instruct-v0.3, mixtral-8x7b-instruct, mixtral-8x22b-instruct, mistral-large, mistral-nemo-12b-instruct, llama3-chatqa-1.5-70b, llama3-chatqa-1.5-8b, nemotron-4-340b-instruct, nemotron-4-340b-reward, usdcode-llama3-70b-instruct, seallm-7b-v2.5, arctic, solar-10.7b-instruct
模型重定向

{
   "yi-large":"01-ai/yi-large",
   "sea-lion-7b-instruct":"aisingapore/sea-lion-7b-instruct",
   "starcoder2-7b":"bigcode/starcoder2-7b",
   "starcoder2-15b":"bigcode/starcoder2-15b",
   "dbrx-instruct":"databricks/dbrx-instruct",
   "deepseek-coder-6.7b-instruct":"deepseek-ai/deepseek-coder-6.7b-instruct",
   "gemma-7b":"google/gemma-7b",
   "gemma-2b":"google/gemma-2b",
   "gemma-2-9b-it":"google/gemma-2-9b-it",
   "gemma-2-27b-it":"google/gemma-2-27b-it",
   "codegemma-1.1-7b":"google/codegemma-1.1-7b",
   "codegemma-7b":"google/codegemma-7b",
   "recurrentgemma-2b":"google/recurrentgemma-2b",
   "granite-34b-code-instruct":"ibm/granite-34b-code-instruct",
   "granite-8b-code-instruct":"ibm/granite-8b-code-instruct",
   "breeze-7b-instruct":"mediatek/breeze-7b-instruct",
   "codellama-70b":"meta/codellama-70b",
   "llama2-70b":"meta/llama2-70b",
   "llama3-8b":"meta/llama3-8b",
   "llama3-70b":"meta/llama3-70b",
   "llama-3.1-8b-instruct":"meta/llama-3.1-8b-instruct",
   "llama-3.1-70b-instruct":"meta/llama-3.1-70b-instruct",
   "llama-3.1-405b-instruct":"meta/llama-3.1-405b-instruct",
   "phi-3-medium-128k-instruct":"microsoft/phi-3-medium-128k-instruct",
   "phi-3-medium-4k-instruct":"microsoft/phi-3-medium-4k-instruct",
   "phi-3-mini-128k-instruct":"microsoft/phi-3-mini-128k-instruct",
   "phi-3-mini-4k-instruct":"microsoft/phi-3-mini-4k-instruct",
   "phi-3-small-128k-instruct":"microsoft/phi-3-small-128k-instruct",
   "phi-3-small-8k-instruct":"microsoft/phi-3-small-8k-instruct",
   "codestral-22b-instruct-v0.1":"mistralai/codestral-22b-instruct-v0.1",
   "mamba-codestral-7b-v0.1":"mistralai/mamba-codestral-7b-v0.1",
   "mistral-7b-instruct":"mistralai/mistral-7b-instruct",
   "mistral-7b-instruct-v0.3":"mistralai/mistral-7b-instruct-v0.3",
   "mixtral-8x7b-instruct":"mistralai/mixtral-8x7b-instruct",
   "mixtral-8x22b-instruct":"mistralai/mixtral-8x22b-instruct",
   "mistral-large":"mistralai/mistral-large",
   "mistral-nemo-12b-instruct":"nv-mistralai/mistral-nemo-12b-instruct",
   "llama3-chatqa-1.5-70b":"nvidia/llama3-chatqa-1.5-70b",
   "llama3-chatqa-1.5-8b":"nvidia/llama3-chatqa-1.5-8b",
   "nemotron-4-340b-instruct":"nvidia/nemotron-4-340b-instruct",
   "nemotron-4-340b-reward":"nvidia/nemotron-4-340b-reward",
   "usdcode-llama3-70b-instruct":"nvidia/usdcode-llama3-70b-instruct",
   "seallm-7b-v2.5":"seallms/seallm-7b-v2.5",
   "arctic":"snowflake/arctic",
   "solar-10.7b-instruct":"upstage/solar-10.7b-instruct"
}

绘图模型

baseurl: 你的Worker地址
key: 复制得到的Key
model: ~~sdxl-lightning,~~stable-diffusion-3-medium,stable-diffusion-xl,sdxl-turbo
模型重定向

{
   "sdxl-lightning":"bytedance/sdxl-lightning",
   "stable-diffusion-3-medium":"stabilityai/stable-diffusion-3-medium",
   "stable-diffusion-xl":"stabilityai/stable-diffusion-xl",
   "sdxl-turbo":"stabilityai/sdxl-turbo"
}

worker.js

const NVIDIA_BASE_URL = "https://ai.api.nvidia.com/v1/genai/";

addEventListener('fetch', event => {
  console.log('Received a fetch event');
  try {
    event.respondWith(handleRequest(event.request));
  } catch (error) {
    console.error('Top-level error handling', error);
    event.respondWith(handleError(error, 'Top-level error handling'));
  }
});

async function handleRequest(request) {
  console.log('Handling request:', request);
  const url = new URL(request.url);
  const path = url.pathname;

  // 路由表
  const routeMap = {
    'OPTIONS': handleCORS,
    'POST': {
      '/v1/chat/completions': handleChatCompletions
    },
    'GET': {
      '/file/': handleFileProxy
    }
  };

  const methodRoutes = routeMap[request.method];
  if (methodRoutes) {
    if (typeof methodRoutes === 'function') {
      console.log('Matching route found for method:', request.method);
      return await methodRoutes(request);
    } else {
      const matchingPath = Object.keys(methodRoutes).find(route => path.startsWith(route));
      if (matchingPath) {
        console.log('Matching path found:', matchingPath);
        return await methodRoutes[matchingPath](request, path);
      }
    }
  }

  // 默认 404 响应
  console.log('No matching route found. Returning 404.');
  return new Response('Not Found', { status: 404 });
}

function handleCORS() {
  console.log('Handling CORS request');
  const headers = {
    'Access-Control-Allow-Origin': '*',
    'Access-Control-Allow-Headers': '*',
    'Access-Control-Allow-Methods': 'GET, POST, OPTIONS'
  };
  return new Response(null, { headers });
}

async function handleFileProxy(request, path) {
  console.log('Handling file proxy request for path:', path);
  const imageUrl = `https://telegra.ph${path}`;
  const ifModifiedSince = request.headers.get('if-modified-since');

  const response = await fetch(imageUrl, {
    headers: {
      'If-Modified-Since': ifModifiedSince
    }
  });

  if (response.status === 304) {
    console.log('File not modified. Returning 304.');
    return new Response(null, {
      status: 304,
      headers: {
        'Last-Modified': ifModifiedSince,
        'Cache-Control': 'public, max-age=31536000'
      }
    });
  }

  const headers = new Headers(response.headers);
  headers.set('Cache-Control', 'public, max-age=31536000');

  console.log('File fetched successfully. Returning file response.');
  return new Response(response.body, {
    status: response.status,
    headers: headers
  });
}

async function handleChatCompletions(request) {
  console.log('Handling chat completions request');
  try {
    const authHeader = request.headers.get('Authorization');
    if (!authHeader?.startsWith('Bearer ')) {
      console.log('Unauthorized request');
      return new Response("Unauthorized", { status: 401 });
    }

    const { messages, model } = await request.json();
    if (!Array.isArray(messages) || messages.length === 0 || !model) {
      console.log('Invalid request format');
      return new Response("Invalid request format", { status: 400 });
    }

    const prompt = messages[messages.length - 1].content;
    if (typeof prompt !== 'string') {
      console.log('Invalid prompt format');
      return new Response("Invalid prompt format", { status: 400 });
    }

    console.log('Generating images with NVIDIA API');
    const imagesBase64 = await generateImagesWithNvidia(prompt, model, authHeader);

    if (!Array.isArray(imagesBase64)) {
      console.error('Unexpected response format from NVIDIA API:', imagesBase64);
      return new Response('Internal Server Error', { status: 500 });
    }

    console.log('Uploading images to Telegraph');
    const uploadPromises = imagesBase64.map(base64 => {
      const uint8Array = Uint8Array.from(atob(base64), c => c.charCodeAt(0));
      const uploadFormData = new FormData();
      uploadFormData.append('file', new Blob([uint8Array], { type: 'image/png' }), 'image.png');

      return fetch('https://telegra.ph/upload', {
        method: 'POST',
        body: uploadFormData
      });
    });

    const uploadResponses = await Promise.all(uploadPromises);

    const workerOrigin = request.url.match(/^(https?:\/\/[^\/]*?)\//)[1];
    const imagesURLs = await Promise.all(uploadResponses.map(async (response) => {
      if (!response.ok) {
        console.error('Failed to upload image');
        throw new Error('Failed to upload image');
      }
      const result = await response.json();
      return workerOrigin + result[0].src;
    }));

    const responsePayload = {
      id: `imggen-${Math.floor(Date.now() / 1000)}`,
      object: 'chat.completion.chunk',
      created: Date.now() / 1000,
      model: 'imagegeneration@006',
      choices: [
        {
          index: 0,
          delta: {
            content: imagesURLs.map(url => `![](${url})`).join('\n\n')
          },
          finish_reason: 'stop'
        }
      ]
    };

    console.log('Returning chat completions response');
    return new Response(
      `data: ${JSON.stringify(responsePayload)}\n\n`,
      {
        status: 200,
        headers: {
          'Access-Control-Allow-Origin': '*',
          'Access-Control-Allow-Headers': '*'
        }
      }
    );
  } catch (error) {
    console.error('Error handling chat completions', error);
    return handleError(error, 'Error handling chat completions');
  }
}

async function generateImagesWithNvidia(prompt, model, authorization) {
  console.log('Generating images with model:', model);
  const { url, payload } = getNvidiaRequestDetails(model, prompt);

  const response = await fetch(url, {
    method: 'POST',
    headers: {
      'Authorization': authorization,
      'Content-Type': 'application/json'
    },
    body: JSON.stringify(payload)
  });

  if (!response.ok) {
    console.error('NVIDIA API Error:', response.status);
    throw new Error(`NVIDIA API Error: ${response.status}`);
  }

  const data = await response.json();
  console.log('Received response from NVIDIA API:', data);

  if (Array.isArray(data.artifacts)) {
    return data.artifacts.map(artifact => artifact.base64);
  } else if (data.image) {
    return [data.image];
  } else {
    throw new Error('Unexpected response format from NVIDIA API');
  }
}

function getNvidiaRequestDetails(model, prompt) {
  console.log('Getting NVIDIA request details for model:', model);
  const configs = {
    'bytedance/sdxl-lightning': {
      url: `${NVIDIA_BASE_URL}bytedance/sdxl-lightning`,
      payload: {
        cfg_scale: 0, clip_guidance_preset: "NONE", height: 1024, width: 1024,
        sampler: "K_EULER_ANCESTRAL", samples: 1, seed: 0, steps: 4,
        style_preset: "none", text_prompts: [{ text: prompt, weight: 1 }]
      }
    },
    'stabilityai/stable-diffusion-3-medium': {
      url: `${NVIDIA_BASE_URL}stabilityai/stable-diffusion-3-medium`,
      payload: {
        aspect_ratio: "1:1", cfg_scale: 5, mode: "text-to-image", model: "sd3",
        negative_prompt: "", output_format: "jpeg", prompt, seed: 0, steps: 50
      }
    },
    'stabilityai/sdxl-turbo': {
      url: `${NVIDIA_BASE_URL}stabilityai/sdxl-turbo`,
      payload: {
        height: 512, width: 512, cfg_scale: 0, clip_guidance_preset: "NONE",
        sampler: "K_EULER_ANCESTRAL", samples: 1, seed: 0, steps: 4,
        style_preset: "none", text_prompts: [{ text: prompt, weight: 1 }]
      }
    },
    'stabilityai/stable-diffusion-xl': {
      url: `${NVIDIA_BASE_URL}stabilityai/stable-diffusion-xl`,
      payload: {
        height: 1024, width: 1024, cfg_scale: 5, clip_guidance_preset: "NONE",
        sampler: "K_EULER_ANCESTRAL", samples: 1, seed: 0, steps: 25,
        style_preset: "none", text_prompts: [{ text: prompt, weight: 1 }]
      }
    }
  };

  const config = configs[model];
  if (!config) {
    console.error('Unsupported model:', model);
    throw new Error(`Unsupported model: ${model}`);
  }
  return config;
}

async function uploadImageToTelegraph(imageBase64) {
  console.log('Uploading image to Telegraph');
  const uint8Array = Uint8Array.from(atob(imageBase64), c => c.charCodeAt(0));
  const uploadFormData = new FormData();
  uploadFormData.append('file', new Blob([uint8Array], { type: 'image/png' }), 'image.png');

  const response = await fetch('https://telegra.ph/upload', {
    method: 'POST',
    body: uploadFormData
  });

  if (!response.ok) {
    console.error('Failed to upload image to Telegraph');
    throw new Error('Failed to upload image to Telegraph');
  }

  const [{ src }] = await response.json();
  console.log('Image uploaded to Telegraph. URL:', src);
  return src;
}

function handleError(error, context) {
  console.error(context, error);
  return new Response('Internal Server Error', { status: 500 });
}

参考链接

unique · 2024 年7 月 30 日 14:54

前排前排，太强啦

xjfkkk · 2024 年7 月 30 日 14:56

666

bcptbtptp · 2024 年7 月 30 日 14:56

太酷辣，佬

civil · 2024 年7 月 30 日 14:56

前排，牛哇

bbb · 2024 年7 月 30 日 14:58

太强啦

Dale · 2024 年7 月 30 日 14:58

令人发指的强

huangbaoleng · 2024 年7 月 30 日 14:58

牛啊佬

lsolol · 2024 年7 月 30 日 14:59

大佬厉害

eggacheb · 2024 年7 月 30 日 15:02

好强，感谢分享

xiaolinkangna · 2024 年7 月 30 日 15:06

这个得玩一下了

xiaowanga · 2024 年7 月 30 日 15:06

这个怎么给他搞成企业，大佬

Coker · 2024 年7 月 30 日 15:14

太强了

heishen · 2024 年7 月 30 日 15:19

R佬能挂上我的python转 openai格式么

snaily · 2024 年7 月 30 日 15:25

感谢r佬分享

gpt22236 · 2024 年7 月 30 日 15:27

感谢大佬

tomtom1982 · 2024 年7 月 30 日 15:34

emmm 问题是我找了一圈也没有发现这个api是怎么收费的。。。free的credit用完后怎么收费？

luckylala · 2024 年7 月 30 日 15:36

感谢分享

greg · 2024 年7 月 30 日 15:37

还是worker部署方便，感谢投喂

Keyman · 2024 年7 月 30 日 15:39

又一次看到大佬的分享，真感激

话题		回复	浏览量
Nvidia的模型调用接口不规范，我来给TA规范规范开发调优	1	107	2024 年8 月 28 日
【第一弹】【更新】用不完，根本用不完，部署cf worker无限免费绘画，可Api，支持多模型切换资源荟萃 ChatGPT , OpenAI , 人工智能	28	1702	2024 年9 月 10 日
【第二弹】用不完，根本用不完，部署cf worker无限免费绘画，可Api，支持多模型切换资源荟萃 ChatGPT , OpenAI , 人工智能	22	1174	2024 年9 月 19 日
【尝鲜】llama 405B Copilot 公车 & cf 自部署资源荟萃 Copilot , CoCopilot , 人工智能	16	804	2024 年9 月 18 日
给Flux-schnell网站加上翻译接口和历史记录资源荟萃人工智能	10	167	2024 年8 月 29 日