diff --git a/router/src/main.rs b/router/src/main.rs index 6c13d912..a8651b67 100644 --- a/router/src/main.rs +++ b/router/src/main.rs @@ -309,7 +309,7 @@ async fn main() -> Result<(), RouterError> { let mut tokenizer = Tokenizer::from_file(filename).ok(); if let Some(tokenizer) = &mut tokenizer { if let Some(class) = &tokenizer_config.tokenizer_class { - if (class == "LlamaTokenizer" || class == "LlamaTokenizerFast") && (tokenizer.get_post_processor().is_none() || tokenizer.get_post_processor().unwrap().added_tokens(false) == 0) { + if (class == "LlamaTokenizer" || class == "LlamaTokenizerFast"){ if let Ok(post_processor) = create_post_processor(tokenizer, &tokenizer_config) { tracing::info!("Overriding LlamaTokenizer with TemplateProcessing to follow python override defined in https://github.com/huggingface/transformers/blob/4aa17d00690b7f82c95bb2949ea57e22c35b4336/src/transformers/models/llama/tokenization_llama_fast.py#L203-L205"); tokenizer.with_post_processor(post_processor);