From e1ee189ad1766977247b04f2e5ecebdda2ccb474 Mon Sep 17 00:00:00 2001 From: Awni Hannun Date: Tue, 11 Feb 2025 15:24:54 -0800 Subject: [PATCH] fix generation evaluations --- llms/mlx_lm/evaluate.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/llms/mlx_lm/evaluate.py b/llms/mlx_lm/evaluate.py index ca5e83bb..2f35ade2 100644 --- a/llms/mlx_lm/evaluate.py +++ b/llms/mlx_lm/evaluate.py @@ -295,7 +295,9 @@ class MLXLM(LM): completions = [] for context, until in tqdm(zip(contexts, untils), total=len(contexts)): - context = self._tokenize(context) + context = self.tokenizer.encode( + context, add_special_tokens=not self.use_chat_template + ) max_tokens = min( self._max_tokens, self.tokenizer.model_max_length - len(context),