I’m trying to use the groq moonshotai/kimi-k2-instruct model. According to the documentation on moonshotai/Kimi-K2-Instruct · Hugging Face this model isnt’ compatible with OpenAI or Gemini tokenizers and it appears to use it’s own custom tokenizer. Is it possible to configure discourse to use a cus…

TL;DR pick the closest tokenizer and set the maximum context to some thousands less to allow for the difference to not affect you.

Thanks. So I decided to enroll the services of ChatGPT, Gemini and Grok to help me decided which tokenizer to use; which would be the closest match to the Kimi Instruct TikToken/BPE tokenizer to generate the most accurate output from the model. I must say modern AI models are fairly representative …

Also worth noting … majority of coding agents these days don’t even bother with an accurate tokenizer like Discourse does. They just estimate at 4 letters per token. cl100k will be plenty fine for the vast majority of use cases on llms with slightly different tokenizers.

Comment utiliser les modèles d'IA avec des tokenizers personnalisés

Soutien

Falco (Falco) Décembre 14, 2025, 12:02 2

TL;DR choisissez le tokenizer le plus proche et définissez le contexte maximum à quelques milliers de moins pour que la différence ne vous affecte pas.

Sujet		Réponses	Vues
Adding Semantic Search feature for our self-hosted discourse site Support ai , ai-search	9	253	Mars 19, 2025
Frustrations on AI spam detector Support spam , ai	9	137	Décembre 21, 2025
Configuring OpenRouter language models Integrations ai	0	1033	Décembre 10, 2024
How to implement Mistral with Embeddings Support related-topics , ai	6	228	Mai 11, 2025
Inquiry About AI Plugin Options Support ai	7	133	Novembre 24, 2025

Comment utiliser les modèles d'IA avec des tokenizers personnalisés

Sujets connexes