|
| 1 | +from pydantic import BaseModel, Field |
| 2 | +from pydantic_ai import Agent |
| 3 | +from datetime import date, datetime, time, timedelta |
| 4 | + |
| 5 | +from django.core.files.base import ContentFile |
| 6 | + |
| 7 | +class TranslatedResult(BaseModel): |
| 8 | + title: str = Field(description="The title of the translated article") |
| 9 | + slug: str = Field( |
| 10 | + description="The URL slug. Do not include the language code. Make it similar to the original URL." |
| 11 | + ) |
| 12 | + description: str = Field( |
| 13 | + description="The description of the translated article. Don't mention that it's translated." |
| 14 | + ) |
| 15 | + author: str = Field(description="The author of the translated article") |
| 16 | + tags: list[str] = Field( |
| 17 | + description="List of Python-related tags inferred from the document." |
| 18 | + ) |
| 19 | + written_date: date = Field(description="The written date of the translated article") |
| 20 | + content: str = Field(description="The content of the translated article") |
| 21 | + |
| 22 | +def translate_rssitem(rss_item_id: int): |
| 23 | + from .models import LLMService, LLMUsage, TranslatedContent, RSSItem |
| 24 | + """ |
| 25 | + Translate an RSS item to Korean using AI and save as TranslatedContent. |
| 26 | + |
| 27 | + Args: |
| 28 | + rss_item: RSSItem instance to translate |
| 29 | + |
| 30 | + Returns: |
| 31 | + TranslatedContent: The created translated content instance |
| 32 | + """ |
| 33 | + rss_item = RSSItem.objects.get(id=rss_item_id) |
| 34 | + |
| 35 | + # Get LLM provider and model |
| 36 | + provider, model = LLMService.get_llm_provider_model() |
| 37 | + if not provider or not model: |
| 38 | + raise ValueError("No available LLM service found") |
| 39 | + |
| 40 | + model_name = f"{provider}:{model}" |
| 41 | + |
| 42 | + # Read the crawled content from the file |
| 43 | + if not rss_item.crawled_content: |
| 44 | + raise ValueError("RSS item has no crawled content") |
| 45 | + |
| 46 | + with rss_item.crawled_content.open('r', encoding='utf-8') as f: |
| 47 | + content = f.read() |
| 48 | + |
| 49 | + # Create AI agent for translation |
| 50 | + agent = Agent( |
| 51 | + model_name, |
| 52 | + output_type=TranslatedResult, |
| 53 | + system_prompt="Translate the following markdown article in full to korean" |
| 54 | + ) |
| 55 | + |
| 56 | + # Run translation |
| 57 | + result = agent.run_sync(content) |
| 58 | + |
| 59 | + # Create TranslatedContent instance |
| 60 | + translated_content = TranslatedContent( |
| 61 | + title=result.output.title, |
| 62 | + slug=result.output.slug, |
| 63 | + description=result.output.description, |
| 64 | + author=result.output.author, |
| 65 | + tags=result.output.tags, |
| 66 | + written_date=result.output.written_date, |
| 67 | + model_name=model_name, |
| 68 | + source_rss_item=rss_item, |
| 69 | + source_url=rss_item.link |
| 70 | + ) |
| 71 | + |
| 72 | + # Save the translated content to a file |
| 73 | + content_file = ContentFile(result.output.content, name=f"{rss_item.id}-translated.md") |
| 74 | + translated_content.content.save(f"{rss_item.id}-translated.md", content_file) |
| 75 | + |
| 76 | + # Save the instance |
| 77 | + translated_content.save() |
| 78 | + |
| 79 | + # Create LLM usage record |
| 80 | + usage = result.usage() |
| 81 | + LLMUsage.objects.create( |
| 82 | + model_name=model_name, |
| 83 | + input_tokens=usage.request_tokens, |
| 84 | + output_tokens=usage.response_tokens, |
| 85 | + total_tokens=usage.total_tokens |
| 86 | + ) |
| 87 | + |
| 88 | + return translated_content |
0 commit comments