llama.cpp
cuda : improve cuda pool efficiency using virtual memory
#4606
Merged

cuda : improve cuda pool efficiency using virtual memory #4606

slaren merged 22 commits into master from sl/cuda-virt-pool
slaren
slaren cuda : improve cuda pool efficiency using virtual memory
0d77fbd7
JohannesGaessler
JohannesGaessler commented on 2023-12-22
slaren fix mixtral
eb223dcd
slaren fix cmake build
bd78dc9a
slaren slaren marked this pull request as ready for review 1 year ago
slaren slaren force pushed 1 year ago
slaren check for vmm support, disable for hip
872408cf
slaren slaren force pushed to 872408cf 1 year ago
slaren
ggerganov
ggerganov approved these changes on 2023-12-23
slaren
JohannesGaessler
JohannesGaessler commented on 2023-12-23
slaren
slaren fix hip build
9452d0d5
slaren clarify granularity
20860dae
JohannesGaessler
JohannesGaessler approved these changes on 2023-12-23
slaren move all caps to g_device_caps
4c0f300a
slaren refactor error checking
545f23d0
slaren slaren force pushed 1 year ago
slaren add cuda_pool_alloc, refactor most pool allocations
110b5055
slaren slaren force pushed to 110b5055 1 year ago
slaren slaren requested a review from ggerganov ggerganov 1 year ago
slaren
slaren fix hip build
b7da1ba0
slaren slaren force pushed to b7da1ba0 1 year ago
slaren CUBLAS_TF32_TENSOR_OP_MATH is not a macro
d8b06c21
ggerganov ggerganov added high priority
ggerganov ggerganov added need feedback
slaren
slaren slaren force pushed 1 year ago
slaren slaren force pushed 1 year ago
slaren more hip crap
9f5ac6d2
slaren slaren force pushed to 9f5ac6d2 1 year ago
slaren llama : fix msvc warnings
5eb62622
slaren ggml : fix msvc warnings
6fe9da0f
slaren minor
d8883623
slaren Merge remote-tracking branch 'origin/master' into sl/cuda-virt-pool
26e97b58
slaren minor
ab6ad5e6
slaren cuda : fallback to CPU on host buffer alloc fail
5acc9e50
JohannesGaessler
JohannesGaessler approved these changes on 2023-12-24
slaren Update ggml-cuda.cu
b9c5a6e7
slaren Update ggml-cuda.cu
3081c4e7
slaren
slaren ensure allocations are always aligned
3ad45fc3
slaren act_size -> actual_size
532cb9b9
ggerganov
ggerganov approved these changes on 2023-12-24
slaren slaren merged 5bf3953d into master 1 year ago
slaren slaren deleted the sl/cuda-virt-pool branch 1 year ago
city96
slaren
city96
slaren
JohannesGaessler
city96
SomeoneSerge
SomeoneSerge commented on 2023-12-25

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone