Initialize the model

The runtime exposes a set of C-style functions. Use ccall to call them from JavaScript:

• LogAddFilePointer() (optional; call before init)
• InitNLM(modelPath) to create a model instance

Note: Logging

Logs are read from the in-memory filesystem (MEMFS) under /logs. Create the directory before enabling logging:

// Create the logs directory
module.FS.mkdir('/logs');

// List log files
const logFiles = module.FS.readdir('/logs');

// Read a log file
const logData = module.FS.readFile('/logs/inference.log');
const logText = new TextDecoder('utf-8').decode(logData);

Initialize the model by calling InitNLM with the model path:

module.ccall('LogAddFilePointer', null, [], []);
const instance = module.ccall('InitNLM', 'number', ['string'], ['/model']);

InitNLM returns a pointer that represents the model instance. You must hold onto this pointer for inference and cleanup.

Other exported functions:

• Delete(instancePtr) to free the model instance.
• Inference(instancePtr, prompt) to run the model.
• OutputNLM(instancePtr) to read the output buffer.
• LogSetSilent(...) for runtime logging control.

Next: Run inference and read output