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

mpt : addendum to changeset:1be89c40 - use "req" parameter of GGUF_GET_KEY macro instead of duplicate code

Browse source
This commit is contained in:
Jan Ploski 2023-10-01 01:48:47 +02:00
parent 26c253eda2
commit df072d2d99
1 changed files with 3 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
llama.cpp
View file

@ -1925,12 +1925,10 @@ static void llm_load_hparams(
} break; } break;
case LLM_ARCH_MPT: case LLM_ARCH_MPT:
{ {
GGUF_GET_KEY(ctx, hparams.f_norm_eps, gguf_get_val_f32, GGUF_TYPE_FLOAT32, true, kv(LLM_KV_ATTENTION_LAYERNORM_EPS));
if (gguf_find_key(ctx, kv(LLM_KV_ATTENTION_CLAMP_KQV).c_str()) >= 0) {
GGUF_GET_KEY(ctx, hparams.f_clamp_kqv, gguf_get_val_f32, GGUF_TYPE_FLOAT32, true, kv(LLM_KV_ATTENTION_CLAMP_KQV));
} else {
hparams.f_clamp_kqv = 0.0f; hparams.f_clamp_kqv = 0.0f;
}
GGUF_GET_KEY(ctx, hparams.f_norm_eps, gguf_get_val_f32, GGUF_TYPE_FLOAT32, true, kv(LLM_KV_ATTENTION_LAYERNORM_EPS));
GGUF_GET_KEY(ctx, hparams.f_clamp_kqv, gguf_get_val_f32, GGUF_TYPE_FLOAT32, false, kv(LLM_KV_ATTENTION_CLAMP_KQV));
GGUF_GET_KEY(ctx, hparams.f_max_alibi_bias, gguf_get_val_f32, GGUF_TYPE_FLOAT32, true, kv(LLM_KV_ATTENTION_MAX_ALIBI_BIAS)); GGUF_GET_KEY(ctx, hparams.f_max_alibi_bias, gguf_get_val_f32, GGUF_TYPE_FLOAT32, true, kv(LLM_KV_ATTENTION_MAX_ALIBI_BIAS));
switch (hparams.n_layer) { switch (hparams.n_layer) {