From f261de9fab4a8d8ca99307b039f95698b3bb3549 Mon Sep 17 00:00:00 2001 From: Aleksandras Kostarevas Date: Tue, 23 Apr 2024 15:21:53 -0400 Subject: [PATCH] Revert n_threads and n_threads_batch change --- native/jni/src/ggml/LanguageModel.cpp | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/native/jni/src/ggml/LanguageModel.cpp b/native/jni/src/ggml/LanguageModel.cpp index 6c06227c2..6837857d5 100644 --- a/native/jni/src/ggml/LanguageModel.cpp +++ b/native/jni/src/ggml/LanguageModel.cpp @@ -52,9 +52,8 @@ LanguageModel *LlamaAdapter::createLanguageModel(const std::string &modelPath) { llama_context_params ctx_params = llama_context_default_params(); ctx_params.n_ctx = LLAMA_CONTEXT_SIZE; - ctx_params.n_threads = 4; - ctx_params.n_threads_batch = 4; - ctx_params.n_batch = 128; + ctx_params.n_threads = 1; + ctx_params.n_threads_batch = 1; adapter->n_batch = ctx_params.n_batch;