EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

ggml-quants, llama : removed excess checks (#7274)

Browse Source
This commit is contained in:
Herman Semenov
2024-05-17 07:08:49 +00:00
committed by GitHub
parent e18bc6aaf3
commit 359cbe3f46
3 changed files with 4 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
llama.cpp
View File

@@ -13904,9 +13904,7 @@ llama_token llama_sample_token_mirostat(struct llama_context * ctx, llama_token_
// Sample the next word X using top-k sampling
llama_sample_top_k(nullptr, candidates, int(k), 1);
if (ctx) {
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
}
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
llama_token X = llama_sample_token(ctx, candidates);
t_start_sample_us = ggml_time_us();
@@ -13920,9 +13918,7 @@ llama_token llama_sample_token_mirostat(struct llama_context * ctx, llama_token_
// Update mu using the learning rate and error
*mu = *mu - eta * e;
if (ctx) {
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
}
ctx->t_sample_us += ggml_time_us() - t_start_sample_us;
return X;
}