Better perplexity for 2- and 3-bit quantization for LLaMA-v2-70B #2807
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Outdated suggestions cannot be applied.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.
In LLaMA-v2-70B eight heads share the same
K
andV
attention tensors, and as a result they are 8X smaller than the attentionQ
tensor. The attentionV
tensor is quite important for generation quality, so it is often quantized with more bits when using k_quants. Given this, we can get a nice improvement in perplexity score (as a measure of generation quality) with negligible increase in quantized model size by quantizing the entire attentionV
tensor with 5 bits when the k_quants logic has decided to quantize it with 3 or 4 bits. The table shows the PPL change for a subset of the k_quants: