feat: add support for MoE offloading #12333
Open
+62
−22
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
This commit introduces a new parameter `num_moe_offload` to the Modelfile, allowing users to offload Mixture-of-Experts (MoE) weights to the CPU to reduce VRAM usage. The `num_moe_offload` parameter can be set to: - A positive integer `N` to offload the first `N` MoE layers. - `-1` to offload all MoE layers. - `0` (default) to disable offloading. This is implemented by passing tensor override rules to the underlying `llama.cpp` library, which already supports this functionality. The documentation for the new parameter has also been updated.
|
Thank you but we want to be able to configure this automatically based on available memory rather than making the user configure it, similar to how the rest of Ollama works. I also don't think it handles multiple GPUs properly. In addition, generally we are not adding new features to the old llama engine. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
This commit introduces a new parameter
num_moe_offloadto the Modelfile, allowing users to offload Mixture-of-Experts (MoE) weights to the CPU to reduce VRAM usage.The
num_moe_offloadparameter can be set to:Nto offload the firstNMoE layers.-1to offload all MoE layers.0(default) to disable offloading.This is implemented by passing tensor override rules to the underlying
llama.cpplibrary, which already supports this functionality. The documentation for the new parameter has also been updated.Try to use Jules to solve the #11772