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

ggml-cuda : use graph allocator (#2684)

Browse source 
use a different function for no_alloc to avoid breaking backwards compat, fixes lora

remove 512 n_batch limit

fixed 2048 batch size

cleanup

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
This commit is contained in:
slaren 2023-08-22 15:25:19 +02:00 • committed by GitHub
parent ef3f333d37
commit 1123f7fbdf
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 92 additions and 228 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml-cuda.h
View file

@ -16,9 +16,14 @@ GGML_API bool ggml_cuda_can_mul_mat(const struct ggml_tensor * src0, const str
GGML_API void ggml_cuda_set_tensor_split(const float * tensor_split);
GGML_API void ggml_cuda_transform_tensor(void * data, struct ggml_tensor * tensor);
GGML_API void ggml_cuda_free_data(struct ggml_tensor * tensor);
GGML_API void ggml_cuda_assign_buffers(struct ggml_tensor * tensor);
GGML_API void ggml_cuda_assign_buffers_no_scratch(struct ggml_tensor * tensor);
GGML_API void ggml_cuda_assign_buffers_force_inplace(struct ggml_tensor * tensor);
GGML_API void ggml_cuda_assign_buffers_no_alloc(struct ggml_tensor * tensor);
GGML_API void ggml_cuda_assign_scratch_offset(struct ggml_tensor * tensor, size_t offset);
GGML_API void ggml_cuda_set_main_device(int main_device);
GGML_API void ggml_cuda_set_mul_mat_q(bool mul_mat_q);
GGML_API void ggml_cuda_set_scratch_size(size_t scratch_size);