Vulkan: Disable multi-add for now#581
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
...until we implement it for Vulkan, else it will run on the CPU and performance of MoE models will be terrible.
Also the Vulkan back-end has the very strange restriction that the number of experts times the number of tokens must be
<= 4096for indirect matrix multiplications (as needed in MoE models). Haven't looked into why this restriction is imposed (as I'm not familiar with the Vulkan back-end at all), so for now just using a very recent PR in mainline to split the indirect matrix multiplication into chunks, where each chunks satisfies the restriction.But this basically means a horrible performance for MoE models. Case in point, with DeepSeek-V2-Lite I'm getting in the range of 1600 t/s PP speed (here and in mainline) vs ~9000 with the
ik_llama.cppCUDA back-end on an RTX-4080.Curious if someone is using the Vullkan back-end in
llama.cppto run DeepSeek-V3/R1 and/or Qwen3-235B-A22B and/or LlaMA-4, etc.