From b9af800edd6d8a46528ddc5f62e79b487b94206d Mon Sep 17 00:00:00 2001 From: RAM Date: Sat, 30 Aug 2025 15:58:18 +0800 Subject: [PATCH] [Optimize] Increase zmq buffer size to prevent apiserver too slowly to consume (#3723) (#3728) Co-authored-by: chenjian <1435317881@qq.com> --- fastdeploy/envs.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fastdeploy/envs.py b/fastdeploy/envs.py index 5551c69f9..41e96c913 100644 --- a/fastdeploy/envs.py +++ b/fastdeploy/envs.py @@ -44,7 +44,7 @@ environment_variables: dict[str, Callable[[], Any]] = { # Whether to use HuggingFace tokenizer. "FD_USE_HF_TOKENIZER": lambda: os.getenv("FD_USE_HF_TOKENIZER", 0), # Set the high watermark (HWM) for receiving data during ZMQ initialization - "FD_ZMQ_SNDHWM": lambda: os.getenv("FD_ZMQ_SNDHWM", 10000), + "FD_ZMQ_SNDHWM": lambda: os.getenv("FD_ZMQ_SNDHWM", 64000), # cache kv quant params directory "FD_CACHE_PARAMS": lambda: os.getenv("FD_CACHE_PARAMS", "none"), # Set attention backend. "NATIVE_ATTN", "APPEND_ATTN"