2024-07-05 17:25:58 +00:00
|
|
|
diff --git a/src/llama.cpp b/src/llama.cpp
|
2024-06-07 06:14:33 +00:00
|
|
|
index 40d2ec2c..f34eb79a 100644
|
2024-07-05 17:25:58 +00:00
|
|
|
--- a/src/llama.cpp
|
|
|
|
|
+++ b/src/llama.cpp
|
2024-06-07 06:14:33 +00:00
|
|
|
@@ -6943,7 +6943,7 @@ static struct ggml_tensor * llm_build_kqv(
|
|
|
|
|
struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
|
|
|
|
|
cb(kq, "kq", il);
|
|
|
|
|
|
|
|
|
|
- if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) {
|
|
|
|
|
+ if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX || model.arch == LLM_ARCH_QWEN2) {
|
|
|
|
|
// for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs
|
|
|
|
|
// ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847
|
|
|
|
|
ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
|
