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

ggml : add ggml_soft_max_ext (#4256)

Browse source 
* metal : implement soft_max_ext

* cuda : implement soft_max_ext

* ggml : implement soft_max_ext (CPU)

* batched-bench : print threads

ggml-ci

* metal : simplify soft_max encoding

ggml-ci

* cuda : use 512 threads for soft_max instead of 32

* ggml : update soft max cpu

* cuda : do warp-based block reduce

* cuda : increase max block size to 1024

* cuda : fix warp reduction initialization of shared mem

* metal : warp-based reduction for soft max kernel

* metal : warp-based reduce for rms_norm

* metal : simplify soft max kernel

ggml-ci

* alloc : fix build with debug
This commit is contained in:
Georgi Gerganov 2023-12-01 10:51:24 +02:00 • committed by GitHub
parent 1d144112c0
commit ef47ec18da
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
8 changed files with 311 additions and 196 deletions
Download patch file Download diff file Expand all files Collapse all files

8
ggml.h
View file

@ -1282,6 +1282,14 @@ extern "C" {
struct ggml_context * ctx,
struct ggml_tensor * a);
// fused soft_max(a*scale + mask)
// mask is optional
GGML_API struct ggml_tensor * ggml_soft_max_ext(
struct ggml_context * ctx,
struct ggml_tensor * a,
struct ggml_tensor * mask,
float scale);
GGML_API struct ggml_tensor * ggml_soft_max_back(
struct ggml_context * ctx,
struct ggml_tensor * a,