Syosetu → Astro Translator

Scrapes chapters from a syosetu.com novel, translates via LM Studio, and writes Astro-compatible markdown files.

Setup

uv sync

Point LM_STUDIO_URL in translate_novel.py to your LM Studio server address.

Usage

Translate with model selection

uv run python translate_novel.py

Without arguments, the script prompts you to select model(s) interactively.

Specify model(s) by CLI argument (config key or model ID):

uv run python translate_novel.py unsloth/qwen3.5-9b@q8_0
uv run python translate_novel.py google/gemma-4-26b-a4b unsloth/qwen3.5-9b@q8_0

Flags

Flag	Description
--section N	Translate only section N (1–14)
--chapter N	Translate only chapter N. If --section is also set, N is the chapter-within-section number; otherwise N is the absolute chapter number
--force	Re-translate existing chapters (skips mtime check)

Examples:

# Section 14 only
uv run python translate_novel.py unsloth/qwen3.5-9b@q8_0 --section 14

# Single chapter, force retranslate
uv run python translate_novel.py unsloth/qwen3.5-9b@q8_0 --section 14 --chapter 4 --force

Post-processing

# Normalize spacing for cross-model comparison
python normalize_translations.py [--dry-run] [--verbose] [--normalize-quotes]

# Extract paragraph-aligned comparison data
python extract_comparison.py [--align {direct,merged,both}]

Migration

python migrate_sections.py

Output structure

output/n5290ea/
├── raw/                        # Scraped Japanese source
├── translated/{model}/         # Translation output by model
└── normalized/{model}/         # Spacing-normalized copies for comparison

Model configuration

Models are configured in MODEL_CONFIGS in translate_novel.py. Each entry defines temperature, sampling parameters, chunk size, and the system/user prompts sent to the LM Studio API. Model metadata (GPU layers, context window) is stored in MODEL_METADATA.

Config keys are matched to LM Studio model IDs via the model_id field. If a config key contains /, it's used as-is in output folder names (e.g., unsloth/qwen3.5-9b@q8_0 → unsloth-qwen3.5-9b@q8_0).

Components

Script	Purpose
translate_novel.py	Main pipeline: scrape + translate + write
normalize_translations.py	Post-process translated files for comparison
extract_comparison.py	Build paragraph-aligned comparison data across models
migrate_sections.py	Migrate files to section-based naming