src/lib/ai/huggingface.js

import { InferenceClient } from "@huggingface/inference";

export async function callHuggingFace({
  model, messages, temperature = 1.0, maxTokens = 2048,
  stream = false, apiKey, onToken,
}) {
  const client = new InferenceClient(apiKey);
  const modelName = model.replace(/^hf:/, "");

  const chatMessages = messages.map((m) => ({ role: m.role, content: m.content }));

  if (stream) {
    let fullText = "";
    const out = await client.chatCompletionStream({
      model: modelName,
      messages: chatMessages,
      temperature: Number(temperature),
      max_tokens: Number(maxTokens),
    });
    for await (const chunk of out) {
      const delta = chunk.choices?.[0]?.delta?.content;
      if (delta) { fullText += delta; onToken?.(delta); }
    }
    return { text: fullText, usage: null };
  }

  const out = await client.chatCompletion({
    model: modelName,
    messages: chatMessages,
    temperature: Number(temperature),
    max_tokens: Number(maxTokens),
  });
  return {
    text: out.choices?.[0]?.message?.content || "",
    usage: {
      promptTokens: out.usage?.prompt_tokens,
      completionTokens: out.usage?.completion_tokens,
    },
  };
}