From df072d2d999bc5924bebbe1ee37cdd810ed343fe Mon Sep 17 00:00:00 2001 From: Jan Ploski Date: Sun, 1 Oct 2023 01:48:47 +0200 Subject: [PATCH] mpt : addendum to changeset:1be89c40 - use "req" parameter of GGUF_GET_KEY macro instead of duplicate code --- llama.cpp | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) diff --git a/llama.cpp b/llama.cpp index 7ea6dbe72..81a014d0a 100644 --- a/llama.cpp +++ b/llama.cpp @@ -1925,12 +1925,10 @@ static void llm_load_hparams( } break; case LLM_ARCH_MPT: { + 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)); - 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; - } + 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)); switch (hparams.n_layer) {