mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-12-25 02:44:36 +00:00
llama : disable context-shift for DeepSeek v2 (#8501)
This commit is contained in:
parent
5e116e8dd5
commit
d65a8361fe
@ -15133,6 +15133,10 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
|
|||||||
|
|
||||||
// apply K-shift if needed
|
// apply K-shift if needed
|
||||||
if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE && lctx.kv_self.has_shift) {
|
if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE && lctx.kv_self.has_shift) {
|
||||||
|
if (lctx.model.arch == LLM_ARCH_DEEPSEEK2) { // not supported due to MLA
|
||||||
|
GGML_ASSERT(false && "Deepseek2 does not support K-shift");
|
||||||
|
}
|
||||||
|
|
||||||
{
|
{
|
||||||
ggml_backend_sched_reset(lctx.sched);
|
ggml_backend_sched_reset(lctx.sched);
|
||||||
|
|
||||||
|
Loading…
Reference in New Issue
Block a user