Prerequisites:
- Python 3.8+
- A C++ compiler (
g++,clang, or MSVC on Windows)
python bootstrap.py
uv run api_wrapper.py
uv run jsonprobe.py
or
uv run concurrent_client.py
uv run api_wrapper.py
navigate to http://127.0.0.1:8080 in ur browser
script new behaviors in web client, sample code operating over logits attached for reference
slap some text in the prototype web client
hit 'probe context slices' to drag the probability mass filter around
look at low-probability items with large differences in log-odds from a full-context query
hit [PROCESS LOGIT] to mutate the logit distribution with this difference
enable [SAMPLE LOGIT] and [AUTOREGRESS] if you want probed context slices
to continually resample a decoding chain biased by your MARKOV MOMENTUM MULTIPLIER
lcpp-binding/
├── models/
├── .gitignore
├── api_wrapper.py <-- 2.
├── bootstrap.py <-- 1.
├── concurrent_clientpy <-- 3.
├── download_model.py
├── pyproject.toml
├── jsonprobe.py <-- 3.
├── README.md
└── uv.lock