The converter's K best paths are rescored with CKIP gpt2-tiny-chinese (GPL-3.0, character-level, ~12M parameters, 22 MB as float16): score = lexicon log-prob + 0.6 x LM log-prob, with the recently committed text as context. Inference is plain Accelerate (Apple Silicon and Intel). Modes off / saver (K=3, 15 ms) / standard (K=10, 40 ms) / quality (K=30, 120 ms) in the input menu. A rerank over its time budget falls back to the lexicon; three in a row pause AI for a minute. With auto-pause on, AI pauses when the Mac is hot or the CPU is busy and drops to saver in Low Power Mode. The model loads in the background and is not loaded at all when off. Evaluation (training/lm/eval_rerank.py): chat 96.6% -> 97.5% chars, Wikipedia 92.9% -> 94.7% chars (about 25-30% fewer wrong characters). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
19 lines
744 B
Bash
Executable File
19 lines
744 B
Bash
Executable File
#!/bin/bash
|
|
# Downloads the CKIP gpt2-tiny-chinese language model (GPL-3.0,
|
|
# https://huggingface.co/ckiplab/gpt2-tiny-chinese) and converts it to build/lm/:
|
|
# lm.bin float16 weights (custom format, see training/lm/export_lm.py)
|
|
# lm-vocab.txt
|
|
# build.sh bundles them into the app when present ("AI 選字"). Safe to re-run.
|
|
set -euo pipefail
|
|
cd "$(dirname "$0")/.."
|
|
|
|
SRC=build/models/ckip-gpt2-tiny
|
|
BASE="https://huggingface.co/ckiplab/gpt2-tiny-chinese/resolve/refs%2Fpr%2F1"
|
|
mkdir -p "$SRC" build/lm
|
|
for f in model.safetensors vocab.txt config.json; do
|
|
[ -s "$SRC/$f" ] || curl -sfL -o "$SRC/$f" "$BASE/$f"
|
|
done
|
|
python3 -I training/lm/export_lm.py "$SRC" build/lm/lm.bin
|
|
cp "$SRC/vocab.txt" build/lm/lm-vocab.txt
|
|
ls -lh build/lm
|