ggml : prevent division by zero in FA CPU op

ggerganov · ggerganov · commit ad83f06ca66b · 2025-10-06T18:50:00.000+03:00
diff --git a/ggml/src/ggml-cpu/ops.cpp b/ggml/src/ggml-cpu/ops.cpp
@@ -8135,7 +8135,7 @@ static void ggml_compute_forward_flash_attn_ext_f16(
         }
 
         // V /= S
-        const float S_inv = 1.0f/S;
+        const float S_inv = S == 0.0f ? 0.0f : 1.0f/S;
         ggml_vec_scale_f32(DV, VKQ32, S_inv);
 
         // dst indices

Original file line number	Diff line number	Diff line change
`@@ -8135,7 +8135,7 @@ static void ggml_compute_forward_flash_attn_ext_f16(`
`8135`	`8135`	`}`
`8136`	`8136`
`8137`	`8137`	`// V /= S`
`8138`		`- const float S_inv = 1.0f/S;`
	`8138`	`+ const float S_inv = S == 0.0f ? 0.0f : 1.0f/S;`
`8139`	`8139`	`ggml_vec_scale_f32(DV, VKQ32, S_inv);`
`8140`	`8140`
`8141`	`8141`	`// dst indices`