| Components | All | New | MacOS | Windows | Linux | iOS | ||||
| Examples | Mac & Win | Server | Client | Guides | Statistic | FMM | Blog | Deprecated | Old | |
Llama.Detokenize
Convert the provided tokens into text (inverse of llama_tokenize()).
| Component | Version | macOS | Windows | Linux | Server | iOS SDK |
| Llama | 16.1 | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes |
MBS( "Llama.Detokenize"; LlamaSession; Tokens { ; remove special; unparse special } ) More
Parameters
| Parameter | Description | Example | Flags |
|---|---|---|---|
| LlamaSession | The session identifier. | $$session | |
| Tokens | A JSON array with numbers. The token values depend on the model. |
[234] | |
| remove special | Allow to remove BOS and EOS tokens if model is configured to do so. Pass 1 to allow or 0 to disallow. Default is 1. |
1 | Optional |
| unparse special | If true, special tokens are rendered in the output. Pass 1 to allow or 0 to disallow. Default is 1. |
1 | Optional |
Result
Returns OK or error.
Description
Convert the provided tokens into text (inverse of llama_tokenize()).See also
Release notes
- Version 16.3
- Fixed memory leaks in Llama.Tokenize and Llama.Detokenize functions.
Created 10th February 2026, last changed 10th February 2026
