From a01ddd9605fe3b01fa8e4c9545c35290ad541afe Mon Sep 17 00:00:00 2001 From: Liangsheng Yin Date: Wed, 7 Aug 2024 01:52:10 -0700 Subject: [PATCH] misc: fix the req_to_token member change (#967) --- python/sglang/srt/managers/tp_worker.py | 4 ++-- python/sglang/srt/mem_cache/memory_pool.py | 1 - 2 files changed, 2 insertions(+), 3 deletions(-) diff --git a/python/sglang/srt/managers/tp_worker.py b/python/sglang/srt/managers/tp_worker.py index cd543da34..8471ad187 100644 --- a/python/sglang/srt/managers/tp_worker.py +++ b/python/sglang/srt/managers/tp_worker.py @@ -289,10 +289,10 @@ class ModelTpServer: "KV cache pool leak detected!" ) - if self.req_to_token_pool.can_use_mem_size != self.req_to_token_pool.size: + if len(self.req_to_token_pool.free_slots) != self.req_to_token_pool.size: warnings.warn( "Warning: " - f"available req slots={self.req_to_token_pool.can_use_mem_size}, " + f"available req slots={len(self.req_to_token_pool.free_slots)}, " f"total slots={self.req_to_token_pool.size}\n" "Memory pool leak detected!" ) diff --git a/python/sglang/srt/mem_cache/memory_pool.py b/python/sglang/srt/mem_cache/memory_pool.py index 9036d73d0..37ce4296d 100644 --- a/python/sglang/srt/mem_cache/memory_pool.py +++ b/python/sglang/srt/mem_cache/memory_pool.py @@ -32,7 +32,6 @@ class ReqToTokenPool: self.req_to_token = torch.empty( (size, max_context_len), dtype=torch.int32, device="cuda" ) - self.can_use_mem_size = size def alloc(self, need_size: int) -> List[int]: if need_size > len(self.free_slots):