DeepseekV4: Add fused hyper-connection ops (#25585)
* dsv4 hc-ops * add missing files; * add cparams * update rpc version * address review comments * address review comments
This commit is contained in:
+3
-2
@@ -1187,9 +1187,10 @@ struct llama_model_deepseek4 : public llama_model_base {
|
||||
float kq_scale,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hc_weighted_sum(
|
||||
ggml_tensor * build_hc_pre(
|
||||
ggml_tensor * x,
|
||||
ggml_tensor * weights) const;
|
||||
ggml_tensor * weights,
|
||||
int il) const;
|
||||
|
||||
ggml_tensor * build_hc_sinkhorn(
|
||||
ggml_tensor * comb,
|
||||
|
||||
Reference in New Issue
Block a user