input var
Prompt template used to extract the name of the destination language from all input variables.
This workflow is designed to automate the translation of markdown files used in HUGO projects, with special attention to preserving the file structure and formatting. The flow ensures that only the relevant text content is translated, while technical elements like front matter, markdown structure, and control characters remain intact. This is particularly useful for teams managing multi-language static sites built with HUGO, and looking to scale content localization while maintaining high quality and consistency.
Purpose and Utility
- Automated Translation: The workflow leverages state-of-the-art language models (OpenAI GPT-4 variants) to provide high-quality translations for markdown files.
- Structure Preservation: It carefully maintains the structure of HUGO markdown files, including front matter in TOML format, markdown headers, and special formatting.
- Selective Translation: The flow is designed to avoid translating field names in front matter or text within HTML tags, focusing only on field values and markdown content.
- Scalable Localization: By automating the translation process, this workflow enables rapid scaling to multiple languages with minimal manual effort.
Key Steps in the Workflow
The workflow consists of several interconnected components. Here’s a step-by-step outline:
| Step | Component | Function |
|---|---|---|
| 1 | Chat Input | Accepts the markdown file to be translated and any required variables (e.g., target language). |
| 2 | Prompt Template (input var) | Extracts the destination language name from input variables for downstream use. |
| 3 | LLM OpenAI (nano) | Uses a lightweight GPT-4 model to process prompts. |
| 4 | Generator (get language name) | Generates the destination language name from the provided variables. |
| 5 | Document Retriever (GetBestTranslation) | Searches for existing best translations or context from internal/document sources. |
| 6 | Prompt Template (Prompt) | Crafts a detailed prompt instructing the LLM on how to translate, with restrictions and examples. |
| 7 | LLM OpenAI (full) | Uses a full-feature GPT-4 model (with large context) to perform the translation. |
| 8 | Generator | Executes the translation using the above prompt and model. |
| 9 | Chat Output | Displays the translated markdown file in the output interface. |
Workflow Logic in Detail
- Input Handling: The user submits a markdown file and specifies the target language. The workflow extracts relevant variables for use in prompts.
- Language Extraction: The first part of the workflow determines the name of the target language from the input, using a lightweight LLM and a custom prompt template.
- Contextual Retrieval: It optionally retrieves existing translations or relevant documentation to provide additional context and ensure translation consistency.
- Translation Prompt Construction: A comprehensive prompt is constructed, detailing formatting rules, translation restrictions, and file structure expectations. An example file structure is given to the model, with strict instructions on what to translate and what to preserve.
- Translation Generation: The main translation is performed using a powerful LLM, ensuring high-quality output while strictly adhering to the formatting and structural requirements.
- Output: The translated markdown file is presented for user review or further automated processing.
Why This Workflow is Useful
- Consistency: Ensures that all translated files follow the strict formatting and structural guidelines required by HUGO projects.
- Efficiency: Substantially reduces the manual effort involved in translating and formatting markdown files for static site generators.
- Scalability: Enables easy scaling to multiple languages and large volumes of content.
- Quality Control: By using both context-aware retrieval and explicit translation instructions, it minimizes errors typical of naive machine translation approaches.
Special Considerations
- Field-Specific Rules: The workflow is careful to only translate field values in the front matter, not the field names or structural elements.
- Formatting Integrity: Control characters like
+ + +and markdown/HTML elements are preserved as required by HUGO and TOML specifications. - Extensibility: The modular approach (with retrievers, prompt templates, and generators) allows for easy adaptation as requirements evolve.
In summary, this workflow provides an end-to-end, reliable, and scalable solution for translating HUGO markdown files, making it highly valuable for organizations managing multilingual static sites or documentation projects.
