From 5985d9178222c8e8ccc862eb4fe6e6cf84c793d2 Mon Sep 17 00:00:00 2001 From: Meng Zhang Date: Tue, 13 Jun 2023 01:13:06 -0700 Subject: [PATCH] fix: use int8_float16 to fix SantaCoder-1B (#237) #236 --- crates/ctranslate2-bindings/src/ctranslate2.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/crates/ctranslate2-bindings/src/ctranslate2.cc b/crates/ctranslate2-bindings/src/ctranslate2.cc index 3b71b19..62dacf3 100644 --- a/crates/ctranslate2-bindings/src/ctranslate2.cc +++ b/crates/ctranslate2-bindings/src/ctranslate2.cc @@ -118,7 +118,7 @@ std::shared_ptr create_engine( if (loader.device == ctranslate2::Device::CPU) { loader.compute_type = ctranslate2::ComputeType::INT8; } else if (loader.device == ctranslate2::Device::CUDA) { - loader.compute_type = ctranslate2::ComputeType::FLOAT16; + loader.compute_type = ctranslate2::ComputeType::INT8_FLOAT16; } if (model_type_str == "AutoModelForCausalLM") {