From 3a63a12624b8b029629e7c53c779d8861b0d5971 Mon Sep 17 00:00:00 2001 From: Charles Zhao Date: Mon, 3 Nov 2025 03:17:22 -0800 Subject: [PATCH] Allow prefill only run by allowing max_prompt_size == seq_len PiperOrigin-RevId: 827415258 --- gemma/gemma.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/gemma/gemma.cc b/gemma/gemma.cc index 7991c35..ae0c6c1 100644 --- a/gemma/gemma.cc +++ b/gemma/gemma.cc @@ -511,7 +511,7 @@ static void GenerateT(const ModelConfig& config, // We use a single divisor, so all sequence lengths must be the same. HWY_ASSERT(qbatch.KV(qi).SeqLen() == seq_len); } - if (max_prompt_size >= seq_len) { + if (max_prompt_size > seq_len) { HWY_ABORT("max_prompt_size = %zu, increase --seq_len to at least that.", max_prompt_size); }