src/services/huggingfaceService.js

// Hugging Face Inference API service
export class HuggingFaceService {
  constructor(token) {
    this.token = token;
  }

  async streamChatCompletion(messages, modelConfig, onChunk, onComplete, onError) {
    try {
      console.log('Starting chat completion with model:', modelConfig.endpoint);
      
      // Use the chat completions endpoint which is more reliable
      const response = await fetch(
        'https://api-inference.huggingface.co/models/' + modelConfig.endpoint,
        {
          method: 'POST',
          headers: {
            'Authorization': `Bearer ${this.token}`,
            'Content-Type': 'application/json',
          },
          body: JSON.stringify({
            inputs: this.formatMessagesForInference(messages),
            parameters: {
              max_new_tokens: 1024,
              temperature: 0.7,
              top_p: 0.9,
              do_sample: true,
              return_full_text: false
            },
            options: {
              wait_for_model: true,
              use_cache: false
            },
            stream: true
          })
        }
      );

      if (!response.ok) {
        const errorText = await response.text();
        console.error('API Error:', response.status, errorText);
        throw new Error(`API error: ${response.status} - ${response.statusText}`);
      }

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

      while (true) {
        const { done, value } = await reader.read();
        if (done) break;

        buffer += decoder.decode(value, { stream: true });
        const lines = buffer.split('\n');
        
        buffer = lines.pop() || '';

        for (const line of lines) {
          if (line.trim() === '') continue;
          
          if (line.startsWith('data: ') && line !== 'data: [DONE]') {
            try {
              const jsonData = line.slice(6);
              if (jsonData.trim()) {
                const data = JSON.parse(jsonData);
                // Handle different response formats
                if (data.token && data.token.text) {
                  onChunk(data.token.text);
                } else if (data.generated_text) {
                  onChunk(data.generated_text);
                } else if (data[0] && data[0].generated_text) {
                  onChunk(data[0].generated_text);
                }
              }
            } catch (e) {
              console.log('Skipping invalid JSON line:', line);
            }
          }
        }
      }

      onComplete();
    } catch (error) {
      console.error('Stream error:', error);
      onError(error.message);
    }
  }

  // Alternative method using chat completion format
  async streamChatCompletionAlt(messages, modelConfig, onChunk, onComplete, onError) {
    try {
      console.log('Using chat completion format with model:', modelConfig.endpoint);
      
      const response = await fetch(
        'https://api-inference.huggingface.co/models/' + modelConfig.endpoint,
        {
          method: 'POST',
          headers: {
            'Authorization': `Bearer ${this.token}`,
            'Content-Type': 'application/json',
          },
          body: JSON.stringify({
            inputs: this.formatChatPrompt(messages),
            parameters: {
              max_new_tokens: 1024,
              temperature: 0.7,
              top_p: 0.9,
              do_sample: true,
              return_full_text: false
            },
            options: {
              wait_for_model: true,
              use_cache: false
            },
            stream: true
          })
        }
      );

      if (!response.ok) {
        const errorText = await response.text();
        console.error('API Error:', response.status, errorText);
        throw new Error(`API error: ${response.status} - ${errorText}`);
      }

      const reader = response.body.getReader();
      const decoder = new TextDecoder();
      let buffer = '';
      let accumulatedText = '';

      while (true) {
        const { done, value } = await reader.read();
        if (done) break;

        buffer += decoder.decode(value, { stream: true });
        const lines = buffer.split('\n');
        
        buffer = lines.pop() || '';

        for (const line of lines) {
          if (line.trim() === '') continue;
          
          if (line.startsWith('data: ') && line !== 'data: [DONE]') {
            try {
              const jsonData = line.slice(6);
              if (jsonData.trim()) {
                const data = JSON.parse(jsonData);
                
                // Extract text from different possible response formats
                let newText = '';
                if (data.token && data.token.text) {
                  newText = data.token.text;
                } else if (data.generated_text) {
                  newText = data.generated_text.replace(accumulatedText, '');
                } else if (data[0] && data[0].generated_text) {
                  newText = data[0].generated_text.replace(accumulatedText, '');
                }
                
                if (newText) {
                  accumulatedText += newText;
                  onChunk(newText);
                }
              }
            } catch (e) {
              console.log('Skipping invalid JSON line:', line);
            }
          }
        }
      }

      onComplete();
    } catch (error) {
      console.error('Stream error:', error);
      onError(error.message);
    }
  }

  // Format messages for inference API
  formatMessagesForInference(messages) {
    if (messages.length === 0) return '';
    
    // For single message, just return the content
    if (messages.length === 1) {
      return messages[0].content;
    }
    
    // For multiple messages, format as conversation
    let conversation = '';
    for (const msg of messages) {
      const role = msg.role === 'user' ? 'Human' : 'Assistant';
      conversation += `${role}: ${msg.content}\n`;
    }
    conversation += 'Assistant: ';
    
    return conversation;
  }

  // Format chat prompt
  formatChatPrompt(messages) {
    if (messages.length === 0) return '';
    
    const lastMessage = messages[messages.length - 1];
    return lastMessage.content;
  }
}