model: mistral le chonk (#31376)

* llama: extend DSA and sdpark to new model

* fix ci
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-10-29 15:43:33 +01:00
1 parent 0ac1a7fd35
commit 78feec062d
23 files changed
+1290 -35

No files matched your search

+5 -1
View File
@@ -6,6 +6,7 @@
#include <algorithm>
#include <cassert>
#include <numeric>
//
// llama_kv_cache_dsa
@@ -45,12 +46,15 @@ llama_kv_cache_dsa::llama_kv_cache_dsa(
hparams_lid.n_embd_head_k_full = model.hparams.indexer_head_size;
hparams_lid.rope_type = LLAMA_ROPE_TYPE_NEOX;
// KV sharing applies to the MLA cache only, each indexer owns its keys
std::iota(hparams_lid.kv_source_impl.begin(), hparams_lid.kv_source_impl.end(), 0);
LLAMA_LOG_INFO("%s: creating indexer KV cache, size = %u cells\n", __func__, kv_size);
kv_lid = std::make_unique<llama_kv_cache>(
model, no_alloc, hparams_lid, type_k, type_v,
v_trans, offload, unified, kv_size, n_seq_max,
n_swa, swa_type, nullptr, filter_lid, reuse, nullptr);
n_swa, swa_type, nullptr, filter_lid, nullptr, nullptr);
}
void llama_kv_cache_dsa::clear(bool data) {