SmolLLM

Greasy Fork is available in English.
LLM utility library
当前为 2025-03-04 提交的版本，查看最新版本。
此脚本不应直接安装。它是供其他脚本使用的外部库，要使用该库请加入元指令 // @require https://update.greasyfork.icu/scripts/528704/1546875/SmolLLM.js
换行
// ==UserScript==
// @name         SmolLLM
// @namespace    http://tampermonkey.net/
// @version      0.1.13
// @description  LLM utility library
// @author       RoCry
// @require https://update.greasyfork.icu/scripts/528703/1546610/SimpleBalancer.js
// @license MIT
// ==/UserScript==

class SmolLLM {
  constructor() {
    if (typeof SimpleBalancer === 'undefined') {
      throw new Error('SimpleBalancer is required for SmolLLM to work');
    }

    this.balancer = new SimpleBalancer();
    this.logger = console;
  }

  /**
   * Prepares request data based on the provider
   * 
   * @param {string} prompt - User prompt
   * @param {string} systemPrompt - System prompt 
   * @param {string} modelName - Model name
   * @param {string} providerName - Provider name (anthropic, openai, gemini)
   * @param {string} baseUrl - API base URL
   * @returns {Object} - {url, data} for the request
   */
  prepareRequestData(prompt, systemPrompt, modelName, providerName, baseUrl) {
    let url, data;

    if (providerName === 'anthropic') {
      url = `${baseUrl}/v1/messages`;
      data = {
        model: modelName,
        max_tokens: 4096,
        messages: [{ role: 'user', content: prompt }],
        stream: true
      };
      if (systemPrompt) {
        data.system = systemPrompt;
      }
    } else if (providerName === 'gemini') {
      url = `${baseUrl}/v1beta/models/${modelName}:streamGenerateContent?alt=sse`;
      data = {
        contents: [{ parts: [{ text: prompt }] }]
      };
      if (systemPrompt) {
        data.system_instruction = { parts: [{ text: systemPrompt }] };
      }
    } else {
      // OpenAI compatible APIs
      const messages = [];
      if (systemPrompt) {
        messages.push({ role: 'system', content: systemPrompt });
      }
      messages.push({ role: 'user', content: prompt });

      data = {
        messages: messages,
        model: modelName,
        stream: true
      };

      // Handle URL based on suffix
      if (baseUrl.endsWith('#')) {
        url = baseUrl.slice(0, -1); // Remove the # and use exact URL
      } else if (baseUrl.endsWith('/')) {
        url = `${baseUrl}chat/completions`; // Skip v1 prefix
      } else {
        url = `${baseUrl}/v1/chat/completions`; // Default pattern
      }
    }

    return { url, data };
  }

  prepareHeaders(providerName, apiKey) {
    const headers = {
      'Content-Type': 'application/json'
    };

    if (providerName === 'anthropic') {
      headers['X-API-Key'] = apiKey;
      headers['Anthropic-Version'] = '2023-06-01';
    } else if (providerName === 'gemini') {
      headers['X-Goog-Api-Key'] = apiKey;
    } else {
      headers['Authorization'] = `Bearer ${apiKey}`;
    }

    return headers;
  }

  /**
   * Process SSE stream data for different providers
   * 
   * @param {string} chunk - Data chunk from SSE
   * @param {string} providerName - Provider name
   * @returns {string|null} - Extracted text content or null
   */
  processStreamChunk(chunk, providerName) {
    if (!chunk || chunk === '[DONE]') return null;

    try {
      this.logger.log(`Processing chunk for ${providerName}:`, chunk);
      
      const data = JSON.parse(chunk);

      if (providerName === 'gemini') {
        const candidates = data.candidates || [];
        if (candidates.length > 0 && candidates[0].content) {
          const parts = candidates[0].content.parts;
          if (parts && parts.length > 0) {
            return parts[0].text || '';
          }
        } else {
          this.logger.log(`No content found in chunk for ${providerName}: ${chunk}`);
          return null;
        }
      } else if (providerName === 'anthropic') {
        // Handle content_block_delta which contains the actual text
        if (data.type === 'content_block_delta') {
          const delta = data.delta || {};
          if (delta.type === 'text_delta' || delta.text) {
            return delta.text || '';
          }
        }
        // Anthropic sends various event types - only some contain text
        return null;
      } else {
        // OpenAI compatible format
        const choice = (data.choices || [{}])[0];
        if (choice.finish_reason !== null && choice.finish_reason !== undefined) {
          return null; // End of generation
        }
        return choice.delta && choice.delta.content ? choice.delta.content : null;
      }
    } catch (e) {
      this.logger.error(`Error parsing chunk: ${e.message}, chunk: ${chunk}`);
      return null;
    }
  }

  /**
   * @returns {Promise<string>} - Full final response text
   */
  async askLLM({
    prompt,
    providerName,
    systemPrompt = '',
    model,
    apiKey,
    baseUrl,
    handler = null,  // handler(delta, fullText)
    timeout = 60000
  }) {
    if (!prompt || !providerName || !model || !apiKey || !baseUrl) {
      throw new Error('Required parameters missing');
    }

    // Use balancer to choose API key and base URL pair
    [apiKey, baseUrl] = this.balancer.choosePair(apiKey, baseUrl);

    const { url, data } = this.prepareRequestData(
      prompt, systemPrompt, model, providerName, baseUrl
    );

    const headers = this.prepareHeaders(providerName, apiKey);

    // Log request info (with masked API key)
    const apiKeyPreview = `${apiKey.slice(0, 5)}...${apiKey.slice(-4)}`;
    this.logger.info(
      `[SmolLLM] Request: ${url} | model=${model} | provider=${providerName} | api_key=${apiKeyPreview} | prompt=${prompt.length}`
    );

    // Create an AbortController for timeout handling
    const controller = new AbortController();
    const timeoutId = setTimeout(() => {
      controller.abort();
    }, timeout);

    try {
      const response = await fetch(url, {
        method: 'POST',
        headers: headers,
        body: JSON.stringify(data),
        signal: controller.signal
      });

      if (!response.ok) {
        throw new Error(`HTTP error ${response.status}: ${await response.text() || 'Unknown error'}`);
      }

      // Handle streaming response
      const reader = response.body.getReader();
      const decoder = new TextDecoder();
      let fullText = '';
      let buffer = '';

      while (true) {
        const { done, value } = await reader.read();
        if (done) break;
        
        const chunk = decoder.decode(value, { stream: true });
        buffer += chunk;
        
        // Process SSE data
        this.processSSEChunks(chunk, providerName, (delta) => {
          if (delta) {
            fullText += delta;
            if (handler) handler(delta, fullText);
          }
        });
      }

      clearTimeout(timeoutId);
      return fullText;
    } catch (error) {
      clearTimeout(timeoutId);
      if (error.name === 'AbortError') {
        throw new Error(`Request timed out after ${timeout}ms`);
      }
      throw error;
    }
  }

  /**
   * Process SSE chunks for different providers
   * 
   * @param {string} text - The SSE text chunk
   * @param {string} providerName - Provider name
   * @param {Function} callback - Callback function for each delta
   */
  processSSEChunks(text, providerName, callback) {
    // Split the input by newlines
    const lines = text.split('\n');
    
    for (let line of lines) {
      line = line.trim();
      if (!line) continue;
      
      // Check for data prefix
      if (line.startsWith('data: ')) {
        const data = line.slice(6).trim();
        
        // Skip [DONE] marker
        if (data === '[DONE]') continue;
        
        // Process the chunk based on provider
        const delta = this.processStreamChunk(data, providerName);
        callback(delta);
      }
    }
  }
}

// Make it available globally
window.SmolLLM = SmolLLM;

// Export for module systems if needed
if (typeof module !== 'undefined') {
  module.exports = SmolLLM;
}