server: MTP layer kv-cache should respect draft type ctk (#23646)
This commit is contained in:
@@ -822,6 +822,8 @@ private:
|
|||||||
auto cparams_dft = common_context_params_to_llama(params_dft);
|
auto cparams_dft = common_context_params_to_llama(params_dft);
|
||||||
if (spec_mtp) {
|
if (spec_mtp) {
|
||||||
cparams_dft.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
|
cparams_dft.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
|
||||||
|
cparams_dft.type_k = params_base.speculative.draft.cache_type_k;
|
||||||
|
cparams_dft.type_v = params_base.speculative.draft.cache_type_v;
|
||||||
}
|
}
|
||||||
cparams_dft.n_rs_seq = 0;
|
cparams_dft.n_rs_seq = 0;
|
||||||
|
|
||||||
@@ -940,6 +942,8 @@ private:
|
|||||||
|
|
||||||
auto cparams_mtp = common_context_params_to_llama(params_base);
|
auto cparams_mtp = common_context_params_to_llama(params_base);
|
||||||
cparams_mtp.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
|
cparams_mtp.ctx_type = LLAMA_CONTEXT_TYPE_MTP;
|
||||||
|
cparams_mtp.type_k = params_base.speculative.draft.cache_type_k;
|
||||||
|
cparams_mtp.type_v = params_base.speculative.draft.cache_type_v;
|
||||||
cparams_mtp.n_rs_seq = 0;
|
cparams_mtp.n_rs_seq = 0;
|
||||||
|
|
||||||
ctx_dft.reset(llama_init_from_model(model_tgt, cparams_mtp));
|
ctx_dft.reset(llama_init_from_model(model_tgt, cparams_mtp));
|
||||||
|
|||||||
Reference in New Issue
Block a user