Mixtral scaling: Reduce perplexity from 4.294 to 4.269 #301
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Outdated suggestions cannot be applied.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 PR reduces the perplexity of Mixtral by introducing scaling of individual experts instead of one scale for all the experts in the MoE block.
To use this in vLLM, GGUF, or other integrations, their code needs to be updated to load the new ScaledMixtralSparseMoeBlock. The only modification needed in the inference code is right before the experts get the
current_hidden_states
:Idea by @Sakits, engineering & implementation by @casper-hansen