remove unnecessary scale

This commit is contained in:
leejet 2026-01-16 23:36:55 +08:00
parent 130ca90b2f
commit 93686089a0

View File

@ -1228,7 +1228,7 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_ext_attention_ext(struct ggml_context
bool diag_mask_inf = false, bool diag_mask_inf = false,
bool skip_reshape = false, bool skip_reshape = false,
bool flash_attn = false, bool flash_attn = false,
float kv_scale = 1.0f / 128.f) { // avoid overflow float kv_scale = 1.0f) { // avoid overflow
int64_t L_q; int64_t L_q;
int64_t L_k; int64_t L_k;
int64_t C; int64_t C;
@ -2184,7 +2184,7 @@ public:
bool bias = true, bool bias = true,
bool force_f32 = false, bool force_f32 = false,
bool force_prec_f32 = false, bool force_prec_f32 = false,
float scale = 1.f / 128.f) float scale = 1.f)
: in_features(in_features), : in_features(in_features),
out_features(out_features), out_features(out_features),
bias(bias), bias(bias),