From 2bfab1bb67460af8b125d94ca4c5ab0040c12cd2 Mon Sep 17 00:00:00 2001 From: Yi-Chia Chen <88210435+ycchen-tw@users.noreply.github.com> Date: Wed, 11 Feb 2026 06:08:34 +0800 Subject: [PATCH] Fix radix cache key to include generated tokens in multi-turn (regression) (#16521) Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> --- python/sglang/srt/mem_cache/radix_cache.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/python/sglang/srt/mem_cache/radix_cache.py b/python/sglang/srt/mem_cache/radix_cache.py index 9dba0b349..a4214dcc1 100644 --- a/python/sglang/srt/mem_cache/radix_cache.py +++ b/python/sglang/srt/mem_cache/radix_cache.py @@ -463,7 +463,7 @@ class RadixCache(BasePrefixCache): ] # Maybe convert to bigram keys for EAGLE - keys = convert_to_bigram_key(req.fill_ids) if self.is_eagle else req.fill_ids + keys = convert_to_bigram_key(token_ids) if self.is_eagle else token_ids keys = self._page_align_keys(keys) values = kv_indices[: len(keys)].to(dtype=torch.int64, copy=True) radix_key = RadixKey(keys, req.extra_key, is_bigram=self.is_eagle) @@ -501,7 +501,7 @@ class RadixCache(BasePrefixCache): ] # Maybe convert to bigram keys for EAGLE - keys = convert_to_bigram_key(req.fill_ids) if self.is_eagle else req.fill_ids + keys = convert_to_bigram_key(token_ids) if self.is_eagle else token_ids keys = self._page_align_keys(keys) values = kv_indices[: len(keys)].to(dtype=torch.int64, copy=True) radix_key = RadixKey(keys, req.extra_key, is_bigram=self.is_eagle)