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

llava : add MobileVLM support (#5132)

Browse source 
* New Feature:
    1. Sum_Rows:
        fix cuda kernel overflow
        fix block shape error when nrows too big
    2. Im2Col:
        Support Batch in cuda
        Support f32 to f32 both in cpu && cuda
    3. DepthWiseConv:
        Support by Im2Col && MulMat
    4. Pool_2d:
        Supoort avg pooling in cuda
    5. HardSigmoid:
        Imp in cuda
    6. HardSwish:
        Imp in cuda

* fix tabs instead of spaces

* code clean

* CUDA POOL2D

* ADD POOL2D test case in test-backend-ops.cpp

* code clean

* fix pool2d_kernel

nits

* fix bug in pool2d kernel

* fix avg pooling, count_include_pad

nits

* test-backend-ops : add more pool_2d tests

* cuda : fix warnings and formatting

* ggml : check types in release builds too in pool_2d

* test-backend-ops : remove f16 pool_2d tests

* cuda : more style fixes

* Add assert in ggml_cuda_op_pool2d

* pool2d float padding fallback

* test-backend-ops : add dst_type to im2col

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
JidongZhang-THU 2024-01-31 21:10:15 +08:00 • committed by GitHub
parent b2b9f025e7
commit 15606309a0
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 421 additions and 41 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml.h
View file

@ -1495,7 +1495,8 @@ extern "C" {
int p1,
int d0,
int d1,
bool is_2D);
bool is_2D,
enum ggml_type dst_type);
GGML_API struct ggml_tensor * ggml_conv_depthwise_2d(
struct ggml_context * ctx,