happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

sampling : zero-initialize input buffers

This commit is contained in:
Georgi Gerganov 2025-12-30 20:12:49 +02:00
parent c5de75989e
commit 791ecb94ff
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/llama-sampling.cpp
View File

@ -1156,6 +1156,8 @@ static bool llama_sampler_dist_backend_init(
// Allocate all tensors from our context to the backend
sctx->inp_buf.reset(ggml_backend_alloc_ctx_tensors_from_buft(sctx->inp_ctx.get(), buft));
ggml_backend_buffer_clear(sctx->inp_buf.get(), 0);
}
const bool res = llama_sampler_backend_support(smpl, buft);
@ -1232,7 +1234,8 @@ static void llama_sampler_dist_backend_set_input(struct llama_sampler * smpl) {
// std::uniform_real_distribution<float> with same rng will produce
// different sequences).
std::uniform_real_distribution<double> dist(0.0f, 1.0f);
const float rnd = dist(sctx->rng);
const float rnd = dist(sctx->rng);
ggml_backend_tensor_set(sctx->inp_uniform, &rnd, 0, sizeof(float));
}
@ -3431,6 +3434,8 @@ static bool llama_sampler_logit_bias_backend_init(
// Allocate all tensors from our context to the backend
sctx->inp_buf.reset(ggml_backend_alloc_ctx_tensors_from_buft(sctx->inp_ctx.get(), buft));
ggml_backend_buffer_clear(sctx->inp_buf.get(), 0);
return true;
}