From 9997047d8a8925bb3557036db5cfaaa33bffa75e Mon Sep 17 00:00:00 2001 From: "Wang, Yi A" Date: Wed, 12 Mar 2025 00:47:02 -0700 Subject: [PATCH] int Signed-off-by: Wang, Yi A --- server/text_generation_server/utils/import_utils.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/server/text_generation_server/utils/import_utils.py b/server/text_generation_server/utils/import_utils.py index d2363a32..35cca9cc 100644 --- a/server/text_generation_server/utils/import_utils.py +++ b/server/text_generation_server/utils/import_utils.py @@ -20,7 +20,9 @@ def get_cuda_free_memory(device, memory_fraction): def get_xpu_free_memory(device, memory_fraction): total_free_memory, total_xpu_memory = torch.xpu.mem_get_info(device) memory_fraction = float(os.getenv("XPU_MEMORY_FRACTION", "1.0")) - free_memory = max(0, total_free_memory - (1 - memory_fraction) * total_xpu_memory) + free_memory = max( + 0, int(total_free_memory - (1 - memory_fraction) * total_xpu_memory) + ) return free_memory