vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-13 02:19:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'upstream' into concedo_experimental

Browse source 
# Conflicts:
#	README.md
#	ci/run.sh
This commit is contained in:
Concedo 2024-05-23 21:29:12 +08:00
commit 9af4907b7f
2 changed files with 16 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

12
llama.cpp
View file

@ -6788,7 +6788,7 @@ static struct ggml_tensor * llm_build_kqv(
cur = ggml_flash_attn_ext(ctx, q, k, v, kq_mask, kq_scale, hparams.f_max_alibi_bias);
if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3) {
if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) {
ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);
}
@ -6797,7 +6797,7 @@ static struct ggml_tensor * llm_build_kqv(
struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
cb(kq, "kq", il);
if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3) {
if (model.arch == LLM_ARCH_PHI2 || model.arch == LLM_ARCH_PHI3 || model.arch == LLM_ARCH_GPTNEOX) {
// for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs
// ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847
ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
@ -17724,6 +17724,14 @@ void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_
ctx->cparams.n_threads_batch = n_threads_batch;
}
uint32_t llama_n_threads(struct llama_context * ctx) {
return ctx->cparams.n_threads;
}
uint32_t llama_n_threads_batch(struct llama_context * ctx) {
return ctx->cparams.n_threads_batch;
}
void llama_set_abort_callback(struct llama_context * ctx, bool (*abort_callback)(void * data), void * abort_callback_data) {
ctx->abort_callback = abort_callback;
ctx->abort_callback_data = abort_callback_data;

6
llama.h
View file

@ -761,6 +761,12 @@ extern "C" {
// n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
LLAMA_API void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_t n_threads_batch);
// Get the number of threads used for generation of a single token.
LLAMA_API uint32_t llama_n_threads(struct llama_context * ctx);
// Get the number of threads used for prompt and batch processing (multiple token).
LLAMA_API uint32_t llama_n_threads_batch(struct llama_context * ctx);
// Set whether to use causal attention or not
// If set to true, the model will only attend to the past tokens
LLAMA_API void llama_set_causal_attn(struct llama_context * ctx, bool causal_attn);