vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Add workaround for kv cache

Browse source
This commit is contained in:
Layl Bongers 2024-04-19 10:06:00 +02:00 • committed by Molly Sophia
parent a0aae8d671
commit a866789603
1 changed files with 6 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

6
src/llama.cpp
View file

@ -5799,6 +5799,12 @@ static void llm_load_hparams(
default: model.type = e_model::MODEL_UNKNOWN; default: model.type = e_model::MODEL_UNKNOWN;
} }
} break; } break;
case LLM_ARCH_RWKV:
{
// TODO: Re-using mamba keys right now, but RWKV isn't state-space
ml.get_key(LLM_KV_SSM_INNER_SIZE, hparams.ssm_d_inner);
ml.get_key(LLM_KV_SSM_STATE_SIZE, hparams.ssm_d_state);
} break;
default: (void)0; default: (void)0;
} }