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

metal : pad n_ctx by 32 (#6177)

Browse source 
* metal : require ne00 >= 128 for mat-mat kernels

ggml-ci

* llama : pad n_ctx by 32

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-03-22 09:36:03 +02:00 • committed by GitHub
parent 59c17f02de
commit 95d576b48e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 14 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
llama.cpp
View file

@ -13044,6 +13044,9 @@ struct llama_context * llama_new_context_with_model(
cparams.rope_freq_base = params.rope_freq_base == 0.0f ? hparams.rope_freq_base_train : params.rope_freq_base;
cparams.rope_freq_scale = params.rope_freq_scale == 0.0f ? hparams.rope_freq_scale_train : params.rope_freq_scale;
// this is necessary due to kv_self.n being padded later during inference
cparams.n_ctx = GGML_PAD(cparams.n_ctx, 32);
// with causal attention, the batch size is limited by the context size
cparams.n_batch = hparams.causal_attn ? std::min(cparams.n_ctx, params.n_batch) : params.n_batch;
cparams.n_ubatch = std::min(cparams.n_batch, params.n_ubatch == 0 ? params.n_batch : params.n_ubatch);