Rewrites AI-generated argumentative/academic essays toward human baseline style informed by CAWSE (M/D bands) LOCNESS, and contrast with DeepSeek-generated counterparts. Ships with a fine-tuned LoRA adapter (9.3 MB) and inference script.
| Component | Path | Notes |
|---|---|---|
| --- | --- | --- |
| Inference script | scripts/inference.py | Entry point — humanize() function or CLI |
| LoRA adapters | assets/adapters/adapters.safetensors.json | 12.3 MB base64 JSON; auto-decoded to binary on first run |
| Pattern weights | data/analysis/weights.json | Corpus-derived, loaded by inference at runtime |
| Decoder | scripts/decode_adapters.py | Reconstructs .safetensors binary from JSON (auto or manual) |
| Installer | scripts/install_deps.sh | One-time: pip install mlx mlx-lm transformers + decode |
| Base model | Qwen/Qwen3-8B-MLX-4bit | Downloaded from HuggingFace on first run (~4.5 GB, cached) |
Requirements: Apple Silicon macOS with Python 3.9+.
bash scripts/install_deps.sh # one-time: installs deps + decodes adapter
python scripts/inference.py --file draft.txt # adapter auto-decodes if not already done
Or from Python:
from scripts.inference import humanize
print(humanize("Your AI-drafted essay text here..."))
When humanizing, address higher-weight rows first. Weights are data-driven from corpus analysis (Mann-Whitney); zero-weight rows were not statistically significant.
| ID | Weight | Category | Pattern |
|---|---|---|---|
| --- | ---: | --- | --- |
| P06_CLICHE_METAPHORS | 0.1358 | vocabulary | Cliche metaphors |
| P15_EM_DASH_OVERKILL | 0.1358 | punctuation | Em dash overkill |
| P21_MARKDOWN_ARTIFACTS | 0.1358 | formatting | Markdown artifacts |
| P23_TEXTBOOK_BOLDING | 0.1358 | formatting | Textbook bolding |
| P12_PRESENT_PARTICIPLE_TAIL | 0.1133 | rhetorical | Present participle tailing |
| P10_RULE_OF_THREES | 0.0806 | rhetorical | Rule of threes |
| P04_AI_VOCABULARY | 0.0621 | vocabulary | AI vocabulary |
| P14_COMPULSIVE_SUMMARIES | 0.0598 | rhetorical | Compulsive summaries |
| P05_EXCESSIVE_ADVERBS | 0.0540 | vocabulary | Excessive adverbs |
| P13_OVER_ATTRIBUTION | 0.0529 | rhetorical | Over-attribution |
| P11_FALSE_RANGES | 0.0341 | rhetorical | False ranges |
| P17_TRANSITION_OVERUSE | 0.0001 | punctuation | Overuse of transition words |
| P01_UNDUE_EMPHASIS | 0.0000 | content | Undue emphasis |
| P02_SUPERFICIAL_ANALYSIS | 0.0000 | content | Superficial analysis |
| P03_REGRESSION_TO_MEAN | 0.0000 | content | Regression to the mean |
| P07_REDUNDANT_MODIFIERS | 0.0000 | vocabulary | Redundant modifiers |
| P08_FILLER_HEDGING | 0.0000 | vocabulary | Filler hedging |
| P09_NEGATIVE_PARALLELISM | 0.0000 | rhetorical | Negative parallelisms |
| P16_EN_DASH_AVOIDANCE | 0.0000 | punctuation | En dash / hyphen misuse for ranges |
| P18_COLLABORATIVE_REGISTER | 0.0000 | register | Collaborative register |
| P19_LETTER_FORMALITY | 0.0000 | register | Letter-style formality |
| P20_INSTRUCTIONAL_CONDESCENSION | 0.0000 | register | Instructional condescension |
| P22_EXCESSIVE_LISTS | 0.0000 | formatting | Excessive bulleted/numbered lists |
| P24_EMOJI_SYMBOL | 0.0000 | formatting | Emoji/symbol injection |
Human Merit / Distinction-range writing in CAWSE often shows variable mean dependency distance (MDD); AI prose may cluster more tightly. When humanizing:
$$ LaTeX unless user explicitly requests LaTeX.For non-Apple-Silicon machines or multi-user deployments, run the optional FastAPI server on a Mac host and connect via HTTP/OpenAPI:
pip install fastapi uvicorn[standard]uvicorn api.main:app --host 0.0.0.0 --port 8765 (set HUMANIZE_API_KEY env var for auth)https:///openapi.json POST /v1/humanize with JSON {"text":"..."} (+ Authorization: Bearer …)See references/hosted_api.md for details.
共 1 个版本